Our Froude-scaling programme (research/aerolog-scaling) tests whether a 1:5 model of a cargo aircraft reproduces the full-size aircraft's dynamics, and it identifies the model's modes by fitting noisy pitch-rate and load-factor records. Earlier steps asked how much noise the fit tolerates. This step ties that noise to a real part, the Bosch BMI088 (the data sheet was retrieved today).
Two findings, both arithmetic on data-sheet figures, not simulation:
- Uncalibrated, the offset swamps the signal. The gyro's zero-rate offset of up to 1 deg/s is 100% of a 1 deg/s pitch-rate rms. The accelerometer's 20 mg zero-g offset is 200% of a 0.010 g load-factor rms. Any identification run assumes the IMU was zeroed first.
- After zeroing, temperature is the floor, not averaging. A 30 s static average leaves under 2 mdeg/s. A 5 K warm-up between zeroing and the doublet moves the gyro by 0.075 deg/s (TCO 0.015 deg/s/K), comparable to its white noise. That is 10% of the signal at a pitch-rate rms of 0.75 deg/s.
What we did not do: simulate the offset inside the fit, or find Allan-variance (bias-instability) figures, which this data sheet does not give. Both are logged as next steps. The accelerometer drift figure is a data-sheet upper bound, and the numbers come from one part, not from measurements on an airframe.
Run it: python3 research/aerolog-scaling/experiments/imu_offset_drift/run.py. The log entry is milestone 9v in research/aerolog-scaling/LOG.md.