Reference

Spec quick reference

Last updated 12 August 2026

A specification tells you which standard applies. The standard tells you what to record. This is the step in between — for each record on a precision-maintenance job, which document asks for it, and what has to be written on it for the record to mean anything a year later.

The project specification governs, not this page. Where the spec is silent, these are the documents it usually points at. Section numbers move between editions, so cite the edition your job is bought to, not the one you remember. Nothing here is a substitute for the OEM manual: the machine’s own documentation outranks every general standard below it, and where the two disagree on a tolerance, the manual wins.

Four documents most precision-maintenance specs point at

Document	What it governs	What it makes you record
API 686 Recommended Practice for Machinery Installation	How a machine gets from the truck to running — foundations, grout, baseplates, piping strain, alignment.	Grout batch and cure conditions, soft foot before alignment, cold alignment targets and the readings shot to them, piping strain checked with the flanges broken.
API 610 Centrifugal Pumps for Petroleum, Petrochemical and Natural Gas Industries	The pump itself — construction, baseplate, and the acceptance limits it ships against.	Vibration at rated conditions, bearing housing readings, seal condition, the performance point actually achieved against the one purchased.
ISO 20816 Mechanical vibration — measurement and evaluation on non-rotating parts (supersedes ISO 10816)	How vibration is measured and which zone the number falls in.	Measurement point, direction, units, machine class, the zone boundary applied, and the load condition it was taken at.
ANSI/ASA S2.75 Shaft alignment methodology (formerly the Part 1 shaft alignment standard)	Alignment tolerance as a function of speed, and the method used to reach it.	As-found and as-left offset and angularity, the method, the instrument, and which tolerance table the verdict came from.

Two more turn up constantly without being “the” spec: ISO 21940 for balance quality grades (the G numbers), and the bolting appendix of whichever pressure or structural code the plant runs to. Neither usually governs the whole job; both get cited in the middle of one.

What each record has to carry

The pattern is the same across all four documents, and it is the reason a record either survives review or does not. Every measurement needs three things beside it: what it was measured with, what it was measured against, and the condition it was measured in. Drop any one and the number stops being evidence.

Record	Measured with	Measured against	Condition
Shaft alignment	Instrument and serial, calibration in date	Named tolerance standard — OEM, plant spec, or an ANSI/ASA S2.75 row	Cold, with the thermal targets it was shot to
Soft foot	Indicator, and the bolt torque the readings were taken at	The tolerance the spec sets, usually 2 mils	All four feet, machine bolted as it will run
Vibration	Analyser and transducer, mounting method	ISO 20816 zone for the machine class, or the OEM limit	Speed and load — a reading at part load is not an acceptance reading
Bolt torque	Wrench or tensioner ID, calibration due date	The torque the OEM or the joint spec calls for	Lubricant used, and the pass number in the pattern
Grout	Product and batch, void check method	The pour and cure window the product sheet sets	Ambient and foundation temperature at the pour
Bearing fit	Micrometer and bore gauge	The OEM fit and internal clearance class	Component temperature — a shaft measured hot reads large

Three that come back marked up

An alignment shot cold to zero

A machine that runs hot does not stay where you left it. A pump handling 300°F product can lift twenty thousandths at the coupling by the time it is at temperature, and an alignment set perfectly to zero cold is an alignment that is twenty thousandths out every hour it runs. The record needs the cold targets on it and it needs to say where they came from — OEM published, calculated from the centreline height and material, or measured on a hot check. Assumed zero is a legitimate answer for a machine that does not grow, and it belongs on the record as those words rather than as a blank.

A torque value with no lubricant

The same bolt at the same torque is a different clamp load dry, oiled, and with anti-seize — by a factor that routinely reaches two. A torque figure on its own is a number somebody read off a wrench, not a statement about the joint. Record the lubricant, the grade, and the pattern, and record the pass: a joint pulled straight to final torque in one pass is not the same joint as one brought up in three, whatever the final reading says.

“Within tolerance” with no standard named

OEM, the plant precision-maintenance spec, and the ANSI/ASA S2.75 tables do not agree, and they are not close at high speed. A verdict with no standard beside it cannot be checked by anyone, and the reviewer’s only options are to take your word for it or send it back. They send it back. Name the document and the row.

The tag is what holds it together

Every record on the job hangs off an equipment tag, and the tag is the only thing that lets a reviewer put the alignment sheet, the vibration baseline, the bolt torque record and the grout report side by side. Two habits break this, both of them quietly:

The tag drifts. P-1201A on the alignment sheet, P1201A on the vibration record, 1201-A on the timesheet. Each is legible to a human and none of them matches, so nothing can be pulled together without somebody reading all three.
The train is recorded as one thing. A motor, a gearbox and a pump are three tags with two couplings between them. Recording an alignment against “the pump” leaves the reader guessing which coupling, and a train is aligned one coupling at a time in the field regardless.

What a reviewer is actually checking

Not whether the numbers are good. Whether the record can be trusted — which is a different question, and it is answered before anyone looks at a value:

Was the instrument in calibration on the day? An out-of-cal laser or torque wrench invalidates every reading it produced, retrospectively. This is the first thing checked because it is the cheapest thing to check and the most expensive to find late.
Is there an as-found? A sheet with only as-left cannot say whether the machine was two thousandths out or forty, and that difference is the hours on the invoice.
Does the acceptance criterion appear beside the reading? Not in a spec on a shelf — on the sheet, next to the number.
Do the records agree with each other? A vibration baseline dated before the alignment that was supposed to fix it is the sort of thing that gets an entire package returned.
Is anything conspicuously absent? Nine machines with soft foot records and one without reads as a machine somebody had trouble with, whether or not that is true.

Permits are law, not standards

Energy isolation, confined space, line break, hot work and critical lifts are regulatory obligations with their own paperwork, and they do not live in API 686 or ISO 20816. They are also the records most likely to be requested by somebody who is not reviewing your technical work at all. Keep them with the job rather than in a separate binder in the site office, and record the isolation point and the stored energy separately — a machine can be locked out and still hold a wound spring, a raised load, a charged accumulator, or a rotor that has not finished coasting.

What this page is not

It is not the standards, and it is not a substitute for reading the ones your job is bought to. It is a map of which document asks for what, written down because the question that actually comes up on site is never “what does ISO 20816 say” — it is “this reading is going on a sheet in ten minutes, what else has to be on it”.

Section numbers, tolerance tables and zone boundaries all move between editions. Where a number matters, take it from the edition named in your specification.

Spec quick reference — millwrightqc.com/spec-reference