· Nimrod Rosenberg · Embedded UAV test pilot
Reading ArduPilot Logs After a Test Flight: Vibration and EKF First
A practical order for reading ArduPilot dataflash logs after a test sortie: vibration and clipping, EKF innovations and lane switches, then attitude, power and events.
Why order matters
An ArduPilot dataflash log from a single sortie contains hundreds of message types. If you open it and start plotting whatever looks interesting, you'll find something, and it might not be the thing that matters. I read logs in the same order every time, starting with the data that tells me whether the rest of the data can be trusted.
That means vibration first, then the navigation filter (EKF), then everything else.
Step 1 — Vibration and clipping
The flight controller's accelerometers feed the attitude estimate and the navigation filter. If they're shaken hard enough, the estimate degrades. If they clip (hit the edge of their measurement range), the information is simply lost.
In ArduPilot, the VIBE message gives you:
- VibeX, VibeY, VibeZ: vibration levels on each axis.
- Clip fields: a running count of accelerometer clipping events for each IMU.
What I look for:
- Clipping that increases at all. I treat any rise as a finding.
- Levels against ArduPilot's published guidance for acceptable vibration.
- Where in the flight it changes. Vibration that rises with throttle points at props and motors. Vibration that appears in a specific phase, such as transition or a particular speed, can point at a resonance in the structure.
- The trend across flights. The same aircraft at the same card a week later should look the same. If Z vibration creeps up across the campaign, something is loosening, wearing or cracking.
- Differences between IMUs. If one IMU reads much worse than another, mounting or location is likely.
Common causes I find: unbalanced or damaged props, motor bearing wear, soft or cracked mounts, a flight controller mount that's too stiff or too soft, and cables tugging on the controller.
Step 2 — The navigation filter (EKF)
ArduPilot's EKF fuses IMU, GPS, barometer, compass and airspeed (and other sensors if fitted) into a position, velocity and attitude estimate. When it's unhappy, the aircraft's idea of where it is drifts from reality.
In EKF3 logs (the XK* messages), I look at:
- Innovation test ratios for velocity, position, height, magnetometer and airspeed. Values approaching or exceeding 1 mean the filter is rejecting, or close to rejecting, that sensor's measurements.
- Lane switching. ArduPilot can run multiple EKF cores and switch the primary. A switch is the filter saying one core disagreed badly enough to stop trusting it. I want to know when and why.
- Height source agreement. Barometer against GPS altitude. Propwash and airflow over the static port can make the baro misbehave in certain flight phases, especially in hover and transition on VTOL aircraft.
- Compass health. Magnetometer innovations that move with throttle usually mean current-induced interference.
I also check the GPS messages for satellite count, HDOP and any glitches, because a GPS problem often shows up first as EKF innovations.
Step 3 — Attitude tracking and control margin
Only now do I look at how well the aircraft flew:
- Desired against actual roll, pitch and yaw (
ATTmessage: DesRoll vs Roll and so on). Persistent error, oscillation or overshoot point at tuning. - Rate controller outputs and, for multirotor phases, the motor outputs (
RCOU). If one motor sits much higher than the others in hover, there's an asymmetry: CG, a motor or ESC issue, or a twisted arm. - Control surface deflections in fixed-wing phases. An aircraft using a lot of trim just to fly straight has a rigging or CG problem.
Step 4 — Power and energy
- Battery voltage and current (
BAT), the voltage sag under load and the consumed capacity at landing. - Power in hover against power in cruise. For quadplanes, this is how you measure what transition actually costs.
- Temperature where it's logged, for motors, ESCs or the battery.
Endurance claims should come from this data, not from the datasheet. There's an older post on battery energy density versus field endurance that goes deeper.
Step 5 — Events, modes and messages
- Mode changes (
MODE) and why they happened: pilot command, failsafe or mission. - Errors and events (
ERR,EV,MSG): failsafes, EKF warnings, GPS glitches, arming checks. - Parameter changes during flight if any were made. Every one should be in the report.
Turning it into a report
Every finding goes into the same format:
- Time: the log timestamp or flight phase.
- Observation: what the data shows, with the message and field name.
- Suspected cause: clearly labelled as a hypothesis.
- Action: inspect, change, re-fly, or carry on with a watch item.
That's the format of the interactive log on the TeamFlight homepage: an event, the channel it shows in, what I think caused it and the next test card. It's synthetic data, but it's the real shape of the report.
Make it repeatable
Over a campaign I script the parts that don't need judgement, such as extracting VIBE, EKF and battery summaries per flight, so every sortie produces the same tables. The judgement goes into the interpretation, not into scrolling through plots. If you're starting on this, there's an older post on early anomaly detection in flight logs and one on vibration analysis in flight testing.
Questions
What vibration level is acceptable in ArduPilot?
ArduPilot's own documentation gives guidance on VIBE levels and treats any accelerometer clipping as a problem. I use that as the baseline, then compare against the same aircraft's earlier flights, because a change matters even when the absolute number looks fine.
Which tools do you use to read logs?
Mission Planner or MAVExplorer for quick looks, and Python scripts for repeatable reports across many flights. The tool matters less than reading in the same order every time.
Can you analyse logs remotely without flying the aircraft?
Yes. A log review is a common first engagement. It's more useful with the parameter file, the test card and the pilot's notes for that flight.