Rotor: Vibration Analyzer

Rotor turns your iPhone into a working vibration analyser. Hold it against a
running motor, pump, fan or gearbox and read the overall velocity in mm/s RMS,
graded A to D against ISO 20816, with the machine class cited.

Then it tells you what is wrong, and shows you why it thinks so.

THE NUMBER IS REAL
Acceleration is integrated to velocity in the frequency domain, line by line,
with a high-pass cutoff. It is never cumulative-summed in the time domain, which
turns sensor bias into a wandering ramp. The graded overall is summed from a
rectangular-windowed spectrum so it stays Parseval-exact instead of
double-counting window leakage. If that means something to you, you already know
why most phone vibration apps are wrong.

VERIFY IT YOURSELF
Rotor ships with a self-test. It runs a signal whose answer is known
analytically through the same chain a real machine goes through, and shows you
measured against expected, side by side.

IT NAMES THE FAULT
Unbalance, misalignment, looseness, blade and vane pass, gear mesh with
sidebands, belt defects and electrical faults at twice line frequency. Every
verdict quotes the lines it came from and gives you the specific next step.
Check the coupling cold and hot. Inspect the impeller for build-up. Cut the
power and watch whether the peak vanishes.

IT KNOWS YOUR MACHINE
Ten presets set the ISO class, the plausible speed range, where to hold the
phone and which faults happen on that kind of kit. Enter the poles and supply
frequency off the nameplate and Rotor computes running speed including slip.
Count the blades or gear teeth, or measure the belt, and it draws those
frequencies as labelled cursors.

THREE AXES AT ONCE
High axial vibration relative to radial is how misalignment is told apart from
unbalance, so Rotor captures all three axes and shows which is worst.

IT REFUSES TO GUESS
If the sensor clipped, Rotor will not grade the capture, because a clipped
waveform invents harmonics that look like severe looseness. If the signal is at
the noise floor it asks whether the machine is running rather than awarding a
confident Zone A to a desk. If a bearing's frequencies sit above what your phone
can sample, it says so, because a missing cursor is not a healthy bearing.

WATCH IT GET WORSE
Log a reading and it keeps the full spectrum. Next month the trend line shows
the overall crossing out of green, with the ISO zone bands drawn behind it.
Rotor also compares against that machine's own baseline and lists the individual
lines that grew, in dB.

BEARINGS, THE HARD WAY
Envelope demodulation, done properly. Band-pass the bearing's resonance, take
the analytic signal by Hilbert transform, transform the envelope, and the defect
rate falls out of the noise. Rotor matches it against the computed BPFO, BPFI,
BSF and FTF to name which part is failing, computed from the bearing geometry.

FIELD BALANCING
Single and two plane by the influence coefficient method. Measure, bolt on a
trial weight, measure again, and Rotor solves the correction weight and angle
on a polar plot.

READ THE SPECTRUM
Orders axis, so 1x and 2x land on integers and a bearing line visibly does not.
Peak table with frequency, order and likely cause. Linear, exponential and
peak-hold averaging. Selectable resolution with the capture time shown.
Hanning, flat-top and uniform windows. mm/s and microns or in/s and mils, with
the ISO boundaries restated in both so the toggle never changes a verdict.

PRIVATE BY DESIGN
Everything runs on the device. No account, no cloud, no analytics, no network
access at all.

BE CLEAR ABOUT THE LIMITS
The built-in accelerometer is usable to about 100 Hz. That covers unbalance,
misalignment, looseness and ISO overall velocity on most slow and medium speed
machines. It cannot see most bearing frequencies, and Rotor says so in the
interface at the moment it matters rather than letting you believe a flat
spectrum means a healthy bearing. That honesty is the reason to trust every
other number in the app.