Summary

Focuses on "sensor failure".

Biomisa Arducopter Sensory Critique (BASiC) dataset for UAV sensor failure. analysis:

• 70 autonomous flight data spanning 7 hours 21 minutes.
• 3 hours 2 minutes of pre-failure and 3 hours 15 mintues of post-failure data.
• 1 hour 4 minutes of no-failure data.
• Raw Data
• Original Dataset

Performed in SITL with parameters to trigger faults.

Diversity of Sensor Failure Scenarios

• GPS -> Shut off
• Remote control -> Shut off
• Accelerometer -> Shut off
• Gyroscope -> Shut off
• Compass -> Shut off
• Barometer -> Shut off

Processing Data

Each message type is represented by a distinct CSV file with enhanced attribute names.

Attribute standardisation introduces "Status" attribute (binary) when any failure occurs during flight.

Data is time-indexed.

Missing values are interpolated.

Columns are structured systematically.

Data Formats

Dataflash binary logs

• On-board memory of Arducopter firmware without preprocessing.
• Mission Planner, MAVExplorer, UAV LogViewer, QGroundControl, or Dronee Plotter.

Dataflash text logs

• Converted to .csv from binary logs.

Processed data

• Time-serialising.
• Data-smooting.
• Linear interpolations to fill missing values for a smooth spatio-temporal dataset.

MATLAB .mat files

Raw data logs (rlogs)

Telemetry logs

Experimental Design, Materials and Methods

Ardupilot firmware modified with parameters that trigger faults in sensor modules.

The method is to disable the sensor altogether with a parameter like

SIM_GPS_DISABLE and setting it to one through MAVProxy.

Logistics:

• 10 different flight locations with unique mission.
• 7 flights within each location with six sensor failures and one control.

Evaluation

SITL means behaviour is model not always real.

Ardupilot was the platform of choice. Others are opined but not attempted.

Individual sensor failures, not combined or multiple.