Battery Life Estimator (UAV) — Lean, Calibrate-and-Go

Lean, calibrate-and-go endurance modeling for multirotors or fixed-wing UAVs.

Start with 5–10 flights, fit a tiny model, and get total and remaining flight time estimates in minutes.

⚡️ Why this exists: Operators need a quick, explainable estimate—not a black box. This MVP is deliberately small and extensible.

What it does

• Calibrate from a CSV of past flights (payload, speed, wind, flight time)
• Estimate total flight time for new payload/speed/wind conditions
• Estimate remaining time using current LiPo per-cell voltage → SOC
• Explainable coefficients you can inspect, version, and compare

How it works (MVP model)

We model the inverse of flight time (minutes⁻¹) as a linear function:

• Headwind is positive (increases airspeed/power); tailwind is negative.
• Remaining time is computed as:

using a conservative LiPo voltage→SOC curve.

Good enough when airframe + mission profile are consistent. You can extend later (hover ratio, density altitude, prop/disk loading, battery aging, interactions).

Quick start

1) Prepare your CSV

Include at least 5–10 lines like:

Columns (required):

• flight_time_min — total flight time (minutes)
• payload_kg — payload mass in kilograms
• ground_speed_mps — ground speed (m/s) during steady cruise
• headwind_mps — positive=headwind, negative=tailwind (m/s)

Tip: Use the steady-cruise portion or mission average. Keep logs consistent across flights.

2) Calibrate the model

python battery_life_estimator.py calibrate   --csv example_flights.csv   --out model.json

Outputs:

• Coefficients: b0, b1, b2
• Fit metrics: MAE (min), MAPE (%)
• Saved model: model.json

3) Estimate total flight time

python battery_life_estimator.py estimate-total   --model model.json   --payload-kg 0.6   --ground-speed-mps 10   --headwind-mps 2

4) Estimate remaining time (with voltage)

python battery_life_estimator.py estimate-remaining   --model model.json   --payload-kg 0.6   --ground-speed-mps 10   --headwind-mps 2   --voltage-per-cell 3.85

LiPo per-cell Voltage → SOC (conservative lookup)

⚠️ Under load, measured voltage sags. Choose a conservative landing threshold (many operators target ~3.5–3.6 V/cell under load). Adjust the curve to match your packs.

When is the MVP “good enough”?

• Your MAE ≤ ~1.5–2.0 min on validation flights of ~15–25 min
• Missions are similar (airframe, props, weather range, and throttle profile)
• You need fast, transparent estimates to aid planning

If errors are larger or conditions vary widely, consider v2 features (below).

Sensible v2 upgrades (when ready)

• Hover ratio (time near hover/throttle band) — biggest gain for multirotors
• Density altitude / air density (from temp & pressure) — seasonal & elevation shifts
• Prop/disk loading (airframe/prop changes) — robustness across configurations
• Battery aging/sag (IR or sag proxy) — stabilize estimates across pack age
• Light non-linear/interaction terms + Ridge (L2) — polish without overfit

Keep it lean: add one or two features only when metrics tell you to.

Calibration tips

• Use consistent cruise segments across flights
• Mix a few payloads and wind conditions to avoid collinearity
• Keep an out-of-sample subset (or use simple K-fold) to validate MAE/MAPE
• Re-calibrate after major changes (new props, airframe, or season)

CLI Reference

calibrate
  --csv <path.csv>
  --out <model.json>

estimate-total
  --model <model.json>
  --payload-kg <float>
  --ground-speed-mps <float>
  --headwind-mps <float>

estimate-remaining
  --model <model.json>
  --payload-kg <float>
  --ground-speed-mps <float>
  --headwind-mps <float>
  --voltage-per-cell <float>

Design principles

• Explainable: linear on inverse time; inspect coefficients directly
• Small data friendly: works with 5–10 quality flights
• Deterministic: no randomness; same inputs → same outputs
• Extensible: drop-in features when metrics require

FAQ

• Does this handle hover-heavy missions?
  The MVP targets steady cruise. If you hover a lot (inspections, photogrammetry), add hover_ratio in v2.
• What about temperature or altitude?
  If errors drift with seasons or elevation, add air density/density altitude.
• Multirotor vs fixed-wing?
  Works for both as a first-order model. Fixed-wing may benefit from a throttle% or aerodynamic term later.
• How often should I recalibrate?
  When coefficients or MAE/MAPE drift after hardware/season changes—or every few months as a routine.

Safety & operational notes

• Always maintain conservative reserves; this is a planning aid, not a hard guarantee.
• Validate on your airframe + packs before mission-critical use.
• Use pack-specific thresholds if aging varies across your fleet.

Roadmap

• v0.2: helpers for density altitude & hover ratio extraction
• v0.3: Ridge option with 1–2 interactions
• v0.4: pack aging proxy (IR/sag) and per-pack calibration

License

MIT — free to use, modify, and embed.

Contact

Questions or ideas? Open an issue or propose a PR in the repo.
Happy (and safe) flying! ✈️