LASC_CanSat — Sanctus Bovis

Descent Analysis Ground Station

   

_{Sanctus Bovis · Magnus Squilla — Orbit NITT}

The SD card comes back before the rocket's story does. Everything downstream of that is arithmetic — raw gyro counts and pressure counts in, a damping ratio and a drag area out. Six stages, fixed order, each one checkable on its own before you trust the next.

No dashboards, no live telemetry, no guessing. Just what actually happened on the way down, reconstructed from what the vehicle logged.

Run

python -m descent.main flight.csv \
       --mass 0.35 --riser 0.50 --diameter 0.30 \
       --bias-file bias_calibration.csv \
       --design-cda 0.15 --out results

python -m descent.main --selftest            # synthetic data, known answers
python -m descent.main --selftest --coning

Stage order

#	module	does
1	load.py	CSV → typed DataFrame, integrity checks
2	preprocess.py	time base → gyro bias → zero-phase filter
3	oscillation.py	PCA → motion class → damping identification
4	aerodynamics.py	density → altitude → velocity → C_D·A → β
5	assessment.py	design intent / consistency / stability
6	plot.py	six figures
7	report.py	assembles it all into one report.html

config.py holds constants and the run configuration. synth.py generates self-test data.

Stage 4 consumes stage 3's output. The C_D·A window is snapped to a whole number of oscillation periods, and C_D·A is corrected for the oscillation bias using the measured modulation. Running the two branches independently is incorrect.

Debugging

Each stage runs on its own — isolate a bad number before it propagates:

python -m descent.load         flight.csv
python -m descent.preprocess   flight.csv --bias-file bias.csv --out pre.csv
python -m descent.oscillation  flight.csv --bias-file bias.csv
python -m descent.aerodynamics flight.csv --no-oscillation

--deploy-ms <int> overrides the deployment instant (a raw t_ms value) on any of main.py, preprocess.py, oscillation.py, or aerodynamics.py, instead of inferring it from the accelerometer. It must fall within the file's t_ms range, snaps to the nearest logged sample, and warns (without stopping) if it leaves under 20% of the record. How deployment was determined — inferred or manual, with any snap distance — is recorded in deployment_method in results.json.

Halt the pipeline mid-way, or keep every stage's output:

python -m descent.main flight.csv --stop-after preprocess
python -m descent.main flight.csv --checkpoint debug/

--checkpoint writes stage_1_load.json … stage_5_assessment.json plus the preprocessed CSV, so a failure in a late stage leaves the earlier ones on disk.

Input files

Flight CSV — required columns: t_ms, gx_dps, gy_dps, gz_dps, pressure_pa, temp_c, rh_pct optional: ax_ms2, ay_ms2, az_ms2, gps_alt_m

t_ms is milliseconds since system boot, not since deployment — it is what --deploy-ms is given in terms of.

Raw pressure_pa is required. Air density cannot be reconstructed from an onboard-derived altitude, and the onboard conversion embeds a fixed standard-atmosphere temperature that this pipeline replaces with the measured one — worth ~8% in C_D·A.

A NaN run up to max_nan_gap_samples (config.py, default 15) is bridged by linear interpolation before anything touches it; longer gaps fail loudly instead of being fed to the filter. Tunable per-run with --max-nan-gap on python -m descent.load standalone. See load.py.

Bias calibration CSV — a few seconds of gyro output with the vehicle still, recorded before flight: gx_dps, gy_dps, gz_dps, temp_c

Report

Every run writes report.html into the output directory, next to the six figures and results.json — one page, no network calls, opens straight in a browser off the SD card if that's all you've got at the site. Regenerate the styling without rerunning the analysis:

python -m descent.report results/

Layout lives in report.py, all styling in report.css — plain CSS, no build step.

Self-test

Generates a flight whose coefficients are chosen in advance and checks what the pipeline returns:

quantity	truth	recovered	error
damping ratio ζ	0.0300	0.0309	2.9%
quadratic coefficient c	0.2000	0.2144	7.2%
natural frequency	0.6200 Hz	0.6202 Hz	0.04%
terminal speed	6.2996 m/s	6.2980 m/s	0.03%
drag area C_D·A	0.1500 m²	0.1498 m²	0.16%

Run it after any change. If Sanctus Bovis's numbers ever look wrong on a real flight, this is the first thing to run — it tells you whether the pipeline broke or the flight was actually weird.

Before real data

• --design-cda from the drogue sizing (criterion 1 is skipped without it)
• bias calibration recorded and its temperature noted
• --mass, --riser, --diameter measured on the recovered vehicle

Rotational energy is always reported as specific energy (J/(kg·m²)), never absolute joules — a single scalar moment of inertia can't correctly represent a non-axisymmetric fuselage, and the swing axis PCA finds is only known after the flight, not a fixed body axis. See oscillation.rotational_energy.

---

_{Orbit NITT · descent analysis for whatever comes down next}