How to Check Shaft Alignment and Work Out Why It Drifted

Why this matters

Setting an alignment and re-checking one are different jobs with different products. Setting it produces a machine inside tolerance. Re-checking it produces a comparison, and the comparison is only worth having if the two readings were taken under conditions you can reduce to each other. A shop that re-aligns a drifting pump every nine months and never asks why has bought a permanent line item; a shop that reads the pattern of the movement usually finds one pipe support and fixes it once. This article is the re-check and the forensic half. Setting an alignment from scratch, and allowing for thermal growth when you align cold, are covered by their own articles and are not re-derived here.

Step 1: isolate, and keep it isolated for the whole check

Open and lock the driver disconnect and prove the terminals dead with a live-dead-live check against a known live source (29 CFR 1910.333(b)(2); NFPA 70E-2021, 120.5). Then work the mechanical side under 29 CFR 1910.147: confirm the rotor has stopped coasting and block it against rotation. Every reading here is taken on a stopped machine. The coupling guard is off for readings and goes back before the machine turns under power, because a guard on mechanical power transmission is required while the machine runs (29 CFR 1910.219). When you turn the shaft to take readings, use a bar on the shaft or a designated barring point with your hands clear of the hub gap, and keep the rotor blocked whenever a hand is inside that gap. A casing that has just run is hot: read it with a non-contact infrared thermometer and treat anything above 140 F as a burn surface.

Skip this and: you are taking dial readings inside a coupling that something can start.

Step 2: record the machine's condition before you record any number

Write down, before the first reading: casing temperature and how long since shutdown, whether the pipe is bolted up or free, whether the hold-down bolts are at torque, and what instrument you are using. That block is not paperwork. It is what makes today's numbers subtractable from the last set.

Skip this and: you have a set of readings that cannot be compared to anything, which is a set of readings that answers no question.

Step 3: clear the prerequisites before you read the coupling

Three things falsify an alignment reading and all three are quicker to check than to chase afterwards.

• Hold-down bolt torque. Loose feet make readings that move between passes.
• Soft foot. A frame that distorts as bolts are torqued makes a coupling reading that describes a machine that stops existing when it is bolted down. Check it before you read the coupling, not after the numbers refuse to converge.
• Coupling and bearing condition. Play in a bearing or a worn element puts slop into every reading. Lever the shaft in its bearing and feel for movement.

Skip these and: you will spend an afternoon correcting a machine that is not misaligned.

Step 4: measure the instrument's own fixed offset

With dial indicators on a bar, bar sag is a fixed systematic offset that acts with gravity, so it lands entirely in the vertical readings and not in the horizontal ones. Measure it: mount the bracket on a length of pipe, rotate it through 180 degrees, and record the reading change. That figure gets subtracted from the vertical rim reading every time.

Because it is a fixed offset from one setup, it does not cancel between two readings taken with two different brackets, which is the trap when a re-check is done by a different tech with different tooling. With a laser system the equivalent offset lives in the mounting hardware and is handled by the instrument's own routine, but the same rule holds: if the two records used different instruments, say so in the file.

Skip this and: every vertical number you record is wrong by the same amount, in the same direction, and it looks perfectly consistent.

Step 5: take both planes, twice

Read vertical offset and vertical angularity, then horizontal offset and horizontal angularity. Take each twice. The spread between the two passes is an independent random figure and it tells you how much of any change you are about to report is real. If the two passes disagree by an amount comparable to the change you are investigating, you have not measured a drift, you have measured your own repeatability, and the answer is to go back to step 3.

Step 6: reduce today's readings to the same condition as the last record

This is the step that turns readings into a comparison, and it is the one that gets skipped. Apply, in order and visibly:

1. Subtract bar sag from the vertical readings.
2. Subtract the recorded thermal offset if today's reading is warm and the reference was cold, or add it if the reverse. The thermal figures come from the original alignment file for this machine, and if they were never recorded, say in your notes that the comparison is uncorrected for temperature rather than pretending it is not needed.
3. Note where a correction is zero and why. Horizontal thermal movement is normally small on a machine that grows symmetrically about its own centerline, so the horizontal correction is usually nil - and writing "nil, symmetric growth" is applying the correction, not skipping it.

Step 7: compare against two things, not one

Compare the corrected numbers against the acceptance figure in the file - which comes from the coupling or machine manufacturer's table and tightens as speed rises, so it is theirs to set - and against the last recorded set. The first says whether the machine is acceptable now. The second says whether something is moving, which is the question that stops the callback.

Step 8: read the pattern of the movement

The reasoning behind the first two rows is worth holding: thermal growth lifts machines, because a machine grows upward from the plane where it is bolted down, and it has very little horizontal authority on a symmetric machine. Pipe has enormous horizontal authority, because a run expanding along its length pushes a nozzle sideways with nothing to stop it but the hold-down bolts. So the plane that moved is your first clue about the family of cause.

Step 9: confirm the suspect with a direct test, not by elimination

If the pattern says pipe strain, prove it: isolate the line on both sides of the pump, lock and tag the valves, relieve the pressure to a gauge reading zero and drain it to a routed drain before a single flange bolt is loosened, because a flange broken loose on a live line sprays its contents at whoever is holding the wrench. Then set an indicator on the coupling hub, loosen the flange bolts, and watch how far the hub moves. A commonly used field acceptance figure is that a hub should not move more than about 2 mils when the flange is broken loose, and the pump manufacturer's allowable nozzle load is the number that actually governs.

Worked example: a pump that was blamed on thermal growth for a year

The record from commissioning, taken cold with pipe bolted: vertical offset 1.5 mils, vertical angularity 0.3 mils per inch, horizontal offset 1.0 mils, horizontal angularity 0.2 mils per inch. The file also records a cold-to-hot vertical growth allowance of 2.8 mils, and a bar sag of 3.0 mils for the bracket used.

The re-check, 14 months later. Taken 40 minutes after shutdown with the casing reading 118 F on a non-contact thermometer, so this is a warm reading, and the condition block says so.

Raw readings, before any correction: vertical offset 7.6 mils, vertical angularity 0.4 mils per inch, horizontal offset 7.5 mils, horizontal angularity 1.9 mils per inch.

Correction 1, bar sag, vertical only: 7.6 - 3.0 = 4.6 mils.

Correction 2, thermal, vertical only: the reference was cold and this reading is warm, so the recorded 2.8 mils of growth comes out: 4.6 - 2.8 = 1.8 mils cold-equivalent vertical offset.

Correction 3, horizontal: sag does not act horizontally and the machine's horizontal thermal movement is recorded as nil for a symmetric growth case, so both corrections are zero and the corrected horizontal offset is the raw 7.5 mils.

The comparison. Vertical went from 1.5 to 1.8 mils, a change of 0.3 mils, which is inside the repeatability spread the two passes showed and is therefore not a finding. Horizontal went from 1.0 to 7.5 mils, seven and a half times the commissioning value, and horizontal angularity went from 0.2 to 1.9 mils per inch, roughly nine and a half times. Against the acceptance figure in the file for this machine at its speed, the corrected horizontal offset is several times over and the vertical is comfortably inside.

What the pattern says. Horizontal moved, vertical did not. Thermal growth on this arrangement acts vertically, so thermal is not the cause even though it had been named as the cause on the two previous work orders. The pattern points at piping.

The confirmation. With the line isolated, locked, relieved to a gauge reading zero and drained, an indicator was set on the coupling hub and the discharge flange bolts were loosened. The hub sprang 11 mils horizontally, against the roughly 2 mils that field practice accepts. That is not an alignment problem with a piping symptom; it is a piping problem with an alignment symptom.

What had actually happened. A hanger two supports downstream had been replaced during unrelated work and set solid instead of on the spring it replaced, so the line's expansion had nowhere to go but into the pump nozzle. Three re-alignments in 14 months, plus a set of bearings, had each faithfully corrected the machine to a pipe that pulled it straight back out. Nobody compared two corrected records; each visit read a machine out of tolerance, put it back in, and closed the ticket. The fix was one hanger.

What getting this wrong looks like. The specific failure is comparing a warm raw reading to a cold reference and reporting the difference as drift. Here that would have given 7.6 against 1.5 vertically, a five-fold apparent vertical movement, which reads exactly like thermal growth or a settling foundation, and would have sent the next tech to shim the feet on a machine whose vertical alignment had not moved at all.

How to verify you got this right

Re-run the check after the fix, at the same condition, and expect the corrected numbers to sit near the commissioning set rather than merely inside tolerance. Then take one more set at a scheduled interval, because a single post-repair reading proves the correction and only a second one proves the cause is gone. File the raw numbers, the corrections you applied, the corrected numbers and the condition block together. A record holding only corrected numbers cannot be re-audited when somebody later disputes the thermal allowance, and one holding only raw numbers is not a comparison at all.

References

• Coupling or machine manufacturer's published alignment tolerance table, which owns the acceptance figure at your machine's speed
• Pump manufacturer's allowable nozzle load figures, which own the pipe strain limit
• 29 CFR 1910.147 for mechanical isolation and blocking the rotor, 29 CFR 1910.333(b)(2) with NFPA 70E-2021, 120.5 for proving the driver dead, and 29 CFR 1910.219 for the coupling guard before the machine runs
• See related: How to Set Up a Shaft Alignment and Know It Held; How to Allow for Thermal Growth When You Align Cold; Why Soft Foot Defeats an Alignment; What Misalignment Does to a Bearing in Mechanical Terms