Step 1 of 4

Choose a recording

Pick a reference recording to see how the analysis behaves, or upload a video of your own machine shot on a phone in slow motion.

Reference recordings

Loading recordings…

Your own recording

Upload a video MP4, MOV, WebM, AVI or MKV, up to 25 MB. Use a tripod if you can; 240 fps slow motion works best. Drop a file here or browse

Step 2 of 4

Machine & analysis setup

Only the capture frame rate is required. Everything else sharpens the result.

Recording

Required

Capture frame rate

fps

The real recording rate. Slow-motion files often report 30 fps in their metadata; enter what the camera actually captured. The highest frequency you can measure is half this value.

Machine details

Optional

Running speed

RPM

Shaft speed from the nameplate or a tachometer.

or expected frequency

Hz

If you know the fault frequency (1× running speed, blade pass…).

Search band from

Hz

to

Hz

Image scale

mm / px

Measure a known length in the frame (e.g. a 100 mm bolt spacing that spans 400 px gives 0.25). Without it, displacement stays in pixels and velocity can't be computed.

Analysis method

Agent · rules Recommended Checks the frame rate, removes camera shake, then zooms in on the fault. Asks for a re-recording when the evidence is weak. Agent · Gemini Same measurement tools; Gemini chooses the next step and writes the summary. Every number still comes from OpenCV. Fixed pipeline Baseline Naive approach for comparison: magnifies the strongest peak anywhere, camera shake included.

Step 3 of 4

Analyzing recording

Each step below is an OpenCV measurement. The agent reads the numbers and decides what to measure next.

Starting… 0.0 s

Activity log

Step 4 of 4

Diagnosis

Original

Magnified fault region

Speed