G-Effects Model by Civil Aerospace Medicine Institute 🚀

Modern, stylish, and professional toolkit for modeling G-induced loss of consciousness (G-LOC) and visualizing aerospace physiology using the FAA’s CGEM foundations. This repository brings together computational models, simulations, and interactive visualizations to support training, safety analysis, and research in aerospace medicine.

Forked from the original FAA AAM-631 CGEM project ("CAMI-Gz-Effects-Model-CGEM-") and extended with modern visualization and configuration tools.

Developed by Dr. Diego Malpica (Direction of Aerospace Medicine, Colombian Aerospace Force, Aerospace Scientific Department). ORCID: 0000-0002-2257-4940.

Highlights

• Physiology-aware modeling 🧬: CGEM-based computations for greyout, blackout, and G-LOC risk.
• Interactive visualizations 📊: Streamlit dashboards for scenarios, thresholds, timelines, and maneuver profiles.
• Reproducible workflows 🔁: Notebooks and scripts for demos and experiments.
• Extensible 🧩: Modular code to customize models, parameters, and data pipelines.

---

Quick Start ⚡

1. Set up the environment

python -m venv .venv
source .venv/bin/activate  # Windows: .venv\Scripts\activate
pip install -r requirements.txt
# Linux users: if CGEM fails to run, install the Fortran runtime
# sudo apt-get update && sudo apt-get install -y libgfortran5

1. Launch the interactive visualization app

streamlit run enhanced_app.py

1. Run the aerobatic profiles demo (optional)

python demo_example.py

1. Explore notebooks

• Open aerobatic_profiles_demo.ipynb or aerobatic_maneuvers_simulation.ipynb in your favorite environment.

TypeScript Frontend (New!)

A modern TypeScript frontend with publication-quality ECharts visualizations is also available:

# Install dependencies
cd frontend
npm install

# Run development server
npm run dev

# Build for production
npm run build

The frontend provides:

• Interactive G-force profile visualization with physiological zones
• CGEM model simulation with configurable pilot parameters
• Scientific dashboard with 6+ ECharts visualizations
• Batch analysis for comparing all aerobatic profiles
• Modern glass-morphism UI optimized for presentations and publications

---

Features ✨

• TypeScript Frontend (NEW - ideal for investor demos and publications)

  • Modern React + TypeScript + Vite application
  • Publication-quality Apache ECharts visualizations
  • Glass-morphism dark theme UI with smooth animations
  • Scientific dashboard with: G-force line charts, physiological heatmaps, risk radar, G distribution histograms, state duration charts, cerebral flow dynamics
  • Interactive pilot configuration with all CGEM parameters
  • Batch analysis for comparing all aerobatic maneuvers
  • All DOI references verifiable for journal publication

• Interactive visualization suite (Streamlit)

  • 2D Plotly charts: G vs time with safety zones; G_eff vs thresholds (greyout, blackout, G-LOC)
  • 3D Plotly trajectory: time × G × G_eff with state coloring
  • Animated timeline playback for physiological response
  • Scientific dashboard (ECharts): Lines, Heatmap (flags), Histogram (G distribution), Radar (summary metrics), Scatter (state-colored), Durations (time-in-state), Flows (F_con/F_vis/F_bo), Banks (consciousness/blackout), HLAP, and 3D (ECharts GL)

• Pilot physiology configuration

  • Standard physiology presets (who=1..6) or fully custom inputs (sex, height, BP ranges, cerebral flow thresholds, heart-response tau, reserve banks)
  • Countermeasures and state: G-suit pressure/coverage, AGSM effectiveness, pressure-breathing, other muscle strain, non-AGSM tensing limit, seat tilt, drug-induced HR delay, dehydration level
  • Result caching per (maneuver × pilot configuration)

• Maneuver library and batch analysis

  • Select any included aerobatic profile; view stats and descriptive analysis
  • Batch run across all maneuvers for comparative metrics and charts

• Centrifuge experiment mode

  • Internal ramp-up/ramp-down experiment driver (G0, Gmax, hold@Gmax, dG/dt up/down)

• CGEM integration

  • Wrapper collects times, G, G_eff, consciousness/vision/blackout flags, time-to-events, flows (F_con/F_vis/F_bo), reserve banks (c_bank/bo_bank), and HLAP series

• Cross-platform research app

  • Streamlit UI; works on Windows, macOS, Linux; Docker recipe included
  • Figures can be downloaded via Plotly/ECharts built-in exporters for reporting

---

Scientific background and model overview

CGEM is a physiology-aware model that predicts visual symptoms, G-induced loss of consciousness (G-LOC), and recovery based on cerebral and retinal perfusion. It tracks reserve “banks” for consciousness and vision, and models the effects of countermeasures (G-suit, AGSM, positive pressure breathing), muscle tensing, seat tilt, drug-induced delays, and dehydration on heart-level mean arterial pressure and effective perfusion. The underlying methods, assumptions, and validation are documented in the FAA Office of Aerospace Medicine technical reports and peer‑reviewed literature listed in References.

• Blood oxygen delivery serves as a proxy for “resource flow” to brain/retina; flow depends on perfusion pressure gradients, vascular resistance, oxygenation fraction, and Gz geometry.
• Heart response: mean arterial pressure ramps toward a max with a time constant after Gz > 1.4; negative Gz produces a transient “push–pull” delay when returning to +Gz.
• Visual states: onset (greyout/red‑out) and blackout reflect intraocular pressure thresholds and retinal perfusion; consciousness tracking uses flow thresholds and reserve banks.

Key sources: FAA OAM technical report describing CGEM’s methods and validation and the CGEM User’s Guide (both cited with DOIs below).

---

Validation and verification (concise summary)

Independent FAA validation compared CGEM predictions versus pooled centrifuge datasets (USN/USAF) and aerobatic profiles:

• Time to G‑LOC vs. onset rate: CGEM tracked pooled data within ~1 SD across 0.05–10 G/s onset rates for relaxed participants without countermeasures.
• Recovery of consciousness: predicted absolute incapacitation durations closely matched pooled experimental findings across offset rates.
• Visual thresholds and countermeasures: predicted greyout/blackout and +G tolerances aligned with cohort means and ranges across gear/technique combinations; aerobatic maneuver symptoms matched expert pilot reports.

See the FAA OAM technical report (DOT/FAA/AM‑23/6; DOI in References) for full tables, figures, and methods.

---

Architecture and code map

• Core model: src/cgem.f compiled Fortran executable (cgem/cgem.exe).
• Python wrapper: cgem_wrapper.py
  • Prepares gloc_inp.dat and optional EGP (jerk profile) files
  • Launches the CGEM executable and parses outputs into rich time series
  • Public API:
    • run_cgem_for_profile(profile_id, config: PilotConfig)
    • run_cgem_centrifuge(g0, gmax, gmaxtime, rampup, rampdown, config)
    • PilotConfig captures subject profile or custom physiology and countermeasures
• TypeScript Frontend: frontend/ (React + Vite + TailwindCSS)
  • src/components/charts/: ECharts visualization components
  • src/pages/: Dashboard pages (Overview, Prediction, Dashboard, Batch, Analysis)
  • src/types/: Full TypeScript type definitions for CGEM data
  • src/utils/: Scientific calculations and physiological thresholds
• Apps and demos: enhanced_app.py, app.py, demo_example.py
• Maneuvers: aerobatic_profiles.py and files in Aerobatics_sample_inputs/
• Docs: FAA user guide and “How it Works” summaries in docs/

---

Usage via Python API (quick view)

Import from cgem_wrapper to run any Aerobatics_sample_inputs/ profile with a PilotConfig. Batch and centrifuge modes are supported; see examples below.

---

Supported Aerobatic Maneuvers 🛩️

All maneuver inputs live in Aerobatics_sample_inputs/ and follow the Nz, duration_ms format. The application currently includes:

Identifier	Source file	Description
hammerhead	hammerhead.txt	Hammerhead (stall-turn): vertical climb, 180° yaw, vertical descent
horizontal_rolling_360	horizontalrolling360.txt	360° aileron roll while maintaining level flight
outside_360	outside360.txt	360° outside loop sustaining −G
outside_inside_vert8	outsideinsidevertical8.txt	Vertical figure-of-eight – outside loop bottom, inside loop top
quarter_down_roll	quarterdownroll.txt	Quarter outside loop followed by a downline snap roll
snap_45deg_down_roll	snap45degdownroll.txt	45° downline with a snap roll
half_vert_roll_neg_pull	halfverticalrollwnegpullout.txt	½ vertical roll ending with a negative G pull-out
triple_push_pull_loop	triple_push_pull_loop.txt	Triple push–pull loop: repeated push (−G) then pull (+G) ×3
triple_push_pull_immelmann	triple_push_pull_immelmann.txt	Triple push–pull Immelmann: push–pull + half-roll repeated ×3
triple_push_pull_split_s	triple_push_pull_split_s.txt	Triple push–pull Split S: three consecutive push–pull Split S entries
high_g_turn	high_g_turn.txt	Sustained high-G level turn with 6–7 G plateau and on/off modulation
loop_standard	loop_standard.txt	Standard loop with 3–5 G pull-up and pull-out phases
immelmann_turn	immelmann_turn.txt	Half-loop to half-roll Immelmann with high +G pull-up
split_s	split_s.txt	Split-S: roll inverted then descending half-loop with high +G pull-out
cuban_eight	cuban_eight.txt	Cuban Eight: two looping segments joined by half-rolls
vertical_eight	vertical_eight.txt	Vertical figure eight with repeated +G exposures and brief −G transitions

Notes:

• Some entries are conceptual/demo profiles intended for physiology and risk visualization rather than flight training guidance. You can add your own maneuvers by dropping a properly formatted file into Aerobatics_sample_inputs/ and updating the mapping in aerobatic_profiles.py.

---

Pilot Configuration (Personalized Physiology) 👨‍✈️

You can now personalize the model with pilot-specific parameters from the UI or programmatically. This enables subject-specific predictions and “what-if” countermeasure exploration.

In the Streamlit apps

• Open either enhanced_app.py or app.py via Streamlit and locate the “Pilot configuration” panel.
• Choose a profile and set parameters:
  • Standard subject profile (who): pick one of 6 standard physiology presets or choose “Custom”.
  • Dehydration level: 0.0–1.0. Applied as a modest reduction in baseline/max BP and normal/max cerebral flow.
  • Countermeasures and state:
    • G-suit max pressure (PSI), suit coverage fraction (0.0–0.7)
    • AGSM effectiveness (0–1), pressure breathing max (mmHg)
    • Pre-test other strain HLAP (mmHg), non-AGSM tensing limit (mmHg)
    • Seat tilt (deg), drug-induced heart-rate response delay (s)
  • If you select “Custom”, additional physiology fields appear:
    • Sex, height (cm)
    • Baseline and max blood pressures (BSP/BDP, MSP/MDP)
    • G tolerance multiplier (gtm) and heart response time constant (beta, s)
    • Consciousness and life reserves (bankcon, banklife, s)

Notes:

• When a standard who profile (1..6) is selected, the model’s internal Subject() routine overrides subject physiology (flows, BP, sex, height). Your countermeasure and state inputs still apply.
• When “Custom” is selected, the app writes your physiology fields directly to the model input (equivalent to who=0).
• The app caches results using both the maneuver and the pilot configuration, so different setups won’t conflict.

Programmatic use 💻

You can configure and run CGEM directly from Python using PilotConfig:

from cgem_wrapper import run_cgem_for_profile, PilotConfig

# Example: standard midrange male with some countermeasures and mild dehydration
cfg = PilotConfig(
    who_profile=2,                 # 1..6 for standard subjects; None for custom
    gsuit_max_psi=5.0,
    gsuit_coverage_fraction=0.35,
    agsm_effectiveness=0.5,
    pbg_max_mmhg=20.0,
    dehydration_level=0.3,
    seat_tilt_deg=10.0,
)

result, tmp_dir = run_cgem_for_profile("hammerhead", config=cfg)
print(result.time_to_greyout_s, result.time_to_blackout_s, result.time_to_gloc_s)

Custom physiology example:

cfg = PilotConfig(
    who_profile=None,             # use custom fields below
    male=1, height_cm=176.0,
    baseline_systolic_bp=118.0, baseline_diastolic_bp=78.0,
    max_systolic_bp=185.0, max_diastolic_bp=90.0,
    g_tolerance_multiplier=1.05, heart_response_tau_s=2.3,
    conbank_s=8.0, lifebank_s=180.0,
    agsm_effectiveness=0.6, pbg_max_mmhg=30.0,
    gsuit_max_psi=6.0, gsuit_coverage_fraction=0.4,
    seat_tilt_deg=15.0, drug_delay_s=0.0,
    dehydration_level=0.2,
)
result, _ = run_cgem_for_profile("outside_360", config=cfg)

Dehydration mapping (heuristic): decreases baseline/max BP and slightly reduces normal/max flow; intended for exploratory use only.

---

Reproducibility and provenance

• Deterministic executable: results are deterministic for a given gloc_inp.dat and EGP profile.
• Environment capture: use the Conda or Docker recipes below for stable runs.
• Temporary outputs: wrappers persist run artifacts under a temp directory; keep these to reproduce plots.
• Cite primary sources (FAA CGEM technical report, user guide, and CGEM software DOI) alongside this repository when publishing results.

---

Contributors & Attribution 🙏

• Original model (FAA CGEM): Developed and maintained within the FAA Civil Aerospace Medical Institute (CAMI), AAM-631. Foundational work by Kyle Copeland (FAA CAMI) and collaborators; see source headers in src/cgem.f and the FAA report cited below.
• This fork and application layer: Dr. Diego Malpica (Direction of Aerospace Medicine, Colombian Aerospace Force, Aerospace Scientific Department). ORCID: 0000-0002-2257-4940.
• Upstream origin: Forked from AAM-631/CAMI-Gz-Effects-Model-CGEM-.

Please retain attribution to the FAA CGEM model and authors in derivative works and cite the original FAA report.

Acknowledgments 💡

This work is built upon and inspired by foundational research conducted within the Federal Aviation Administration (FAA) Office of Aerospace Medicine and decades of operational physiology experience. We gratefully acknowledge the contributions of the U.S. Military community—aviators, aircrew, and allied professionals—who served both as scientists and as research participants in the studies underpinning this modeling approach. Their service and commitment to safety and science made this work possible.

Special recognition is due to the FAA researchers and collaborators whose efforts developed, validated, and documented the CGEM approach and related physiology insights.

---

How to cite 📝

When publishing results derived from this code and the CGEM model, please cite:

• Copeland, K., & Whinnery, J. E. (2023). Cerebral blood flow‑based computer modeling of Gz‑induced effects (DOT/FAA/AM‑23/6). Office of Aerospace Medicine, FAA. DOI: https://doi.org/10.21949/1524446
• Copeland, K. (2021). CGEM User’s Guide (DOT/FAA/AM‑23/5). Office of Aerospace Medicine, FAA. DOI: https://doi.org/10.21949/1524438
• CGEM software release (archived package). DOI: https://doi.org/10.21949/1524439

Optionally add a software citation for this repository (include commit hash or release tag) and the specific version of the CGEM executable used.

---

Disclaimer ⚠️

• This toolkit is intended for research, education, and training support. It does not substitute for operational aeromedical guidance or certification processes.
• This project is not an official product of the FAA or the U.S. Department of Defense. All views expressed are those of the contributors.

---

Repository Guide 📁

• enhanced_app.py: Streamlit UI for interactive modeling and visualization
• app.py, demo_example.py: Additional demos and app entry points
• src/: Core model code (e.g., CGEM implementation and related utilities)
• Aerobatics_sample_inputs/: Example input profiles for scenarios
• docs/: Guides and related documents
• notebooks/: Research and demo notebooks

---

Running with Conda (recommended for science stacks) 🧪

Create an isolated Conda environment with all dependencies (CPU-only):

# Create environment
conda create -n cgem-env -y python=3.11

# Activate
conda activate cgem-env

# Core scientific stack
conda install -y -c conda-forge \
  numpy>=1.24 \
  pandas>=2.0 \
  scipy>=1.10 \
  matplotlib>=3.7 \
  seaborn>=0.12 \
  plotly>=5.17 \
  pillow>=10.0 \
  pip

# Streamlit and extras via pip (conda-forge streamlit is OK too)
pip install "streamlit>=1.28"

Run the app:

streamlit run enhanced_app.py

Note:

• The CGEM Fortran executable (cgem) requires the GNU Fortran runtime. On Linux, ensure libgfortran5 is installed (e.g., sudo apt-get install -y libgfortran5). Inside Conda environments this is typically resolved by the system’s shared libraries.

---

Dockerization 🐳

Use Docker to run the app with a reproducible environment.

1) Build the image

Create a Dockerfile in the project root with the following content:

# Base image with Python and build tools
FROM python:3.11-slim

# Install system dependencies (GNU Fortran runtime for CGEM)
RUN apt-get update -y && \
    apt-get install -y --no-install-recommends \
      libgfortran5 \
      && rm -rf /var/lib/apt/lists/*

# Set workdir
WORKDIR /app

# Copy dependency manifests first (leverage Docker layer caching)
COPY requirements.txt ./

# Install Python deps
RUN pip install --no-cache-dir -r requirements.txt

# Copy the rest of the repo
COPY . .

# Streamlit configuration (optional)
ENV STREAMLIT_SERVER_HEADLESS=true \
    STREAMLIT_BROWSER_GATHER_USAGE_STATS=false

# Expose Streamlit default port
EXPOSE 8501

# Default command runs the Streamlit app
CMD ["streamlit", "run", "enhanced_app.py", "--server.port=8501", "--server.address=0.0.0.0"]

Then build:

docker build -t cgem-app:latest .

2) Run the container

docker run --rm -p 8501:8501 cgem-app:latest

Open the app at http://localhost:8501.

3) Development mounts (optional)

To iterate on code without rebuilding:

docker run --rm -p 8501:8501 \
  -v $(pwd):/app \
  cgem-app:latest

This mounts your working directory into the container.

---

Troubleshooting 🛠️

• Missing libgfortran.so.5 when running CGEM:
  • On Debian/Ubuntu: sudo apt-get update && sudo apt-get install -y libgfortran5
• Streamlit not installed inside your environment:
  • Recreate your environment and re-run pip install -r requirements.txt (or Conda steps above).
• Persisting CGEM temp files:
  • The wrapper now stores run artifacts under /tmp/cgem_run_* and returns the path for inspection.

---

References (key sources with DOIs/URLs)

• Besch, E. L., Werchan, P. M., Wiegman, J. F., Nesthus, T. E., & Shahed, A. R. (1994). Effect of hypoxia and hyperoxia on human +Gz duration tolerance. Journal of Applied Physiology, 76(4), 1693–1700. DOI: https://doi.org/10.1152/jappl.1994.76.4.1693
• Copeland, K., & Whinnery, J. E. (2023). Cerebral blood flow‑based computer modeling of Gz‑induced effects (DOT/FAA/AM‑23/6). Office of Aerospace Medicine, FAA. DOI: https://doi.org/10.21949/1524446
• Copeland, K. (2021). CGEM User’s Guide (DOT/FAA/AM‑23/5). Office of Aerospace Medicine, FAA. DOI: https://doi.org/10.21949/1524438
• CGEM software (archived package). DOI: https://doi.org/10.21949/1524439
• Eiken, O., & Grönkvist, M. (2013). Signs and symptoms during supra‑tolerance +Gz exposures, with reference to G‑garment failure. Aviation, Space, and Environmental Medicine, 84(3), 196–205. DOI: https://doi.org/10.3357/asem.3436.2013
• Långsjö, J. W., Alkire, M. T., Kaskinoro, K., et al. (2012). Returning from oblivion: imaging the neural core of consciousness. The Journal of Neuroscience, 32(14), 4935–4943. DOI: https://doi.org/10.1523/JNEUROSCI.4962-11.2012
• Quarry, V. M., & Spodick, D. H. (1974). Cardiac responses to isometric exercise: comparative effects of different postures and levels of exertion. Circulation, 49(5), 905–920. DOI: https://doi.org/10.1161/01.CIR.49.5.905
• Rossen, R., Kabat, H., & Anderson, J. P. (1943). Acute arrest of cerebral circulation in man. Archives of Neurology and Psychiatry, 50(5), 510–528. DOI: https://doi.org/10.1001/archneurpsyc.1943.02290230022002
• Ryoo, H. C., Sun, H. H., Shender, B. S., & Hrebien, L. (2004). Consciousness monitoring using NIRS during high +Gz exposures. Medical Engineering & Physics, 26(9), 745–753. DOI: https://doi.org/10.1016/j.medengphy.2004.07.003
• Sabbahi, A., Arena, R., Kaminsky, L. A., Myers, J., & Phillips, S. A. (2018). Peak blood pressure responses during maximum cardiopulmonary exercise testing: FRIEND reference standards. Hypertension, 71(2), 229–236. DOI: https://doi.org/10.1161/HYPERTENSIONAHA.117.10116
• Tripp, L. D., Warm, J. S., Matthews, G., Chiu, P. Y., & Bracken, R. B. (2009). Cerebral oxygen saturation and pilot performance during G‑LOC. Human Factors, 51(6), 775–784. DOI: https://doi.org/10.1177/0018720809359631
• Whinnery, T., & Forster, E. M. (2015). Neurologic state transitions in the eye and brain: kinetics of loss and recovery of vision and consciousness. Visual Neuroscience, 32, E008. DOI: https://doi.org/10.1017/S095252381500005X
• Whinnery, T., Forster, E. M., & Rogers, P. B. (2014). The +Gz recovery of consciousness curve. Extreme Physiology & Medicine, 3, 9. DOI: https://doi.org/10.1186/2046-7648-3-9

For historical FAA technical reports and broader catalog access, see the FAA Office of Aerospace Medicine portal (https://www.faa.gov/go/oamtechreports) and the National Transportation Library ROSA P repository (https://rosap.ntl.bts.gov).

---

Research: HRV-Based G-LOC Prediction Enhancement 🔬

Executive Summary

This section presents research findings on integrating Heart Rate Variability (HRV) monitoring into the CGEM model for improved G-LOC prediction. Real-time HRV data from wearable devices (e.g., Polar H10 chest straps) offers significant potential for individualized, predictive G-LOC prevention in operational military aviation contexts.

Scientific Background: HRV and Autonomic Nervous System

Heart Rate Variability reflects the beat-to-beat fluctuations in heart rate, governed by the dynamic interplay between the sympathetic (fight-or-flight) and parasympathetic (rest-digest) branches of the autonomic nervous system (ANS). Under high-G stress, the cardiovascular system undergoes profound changes that are detectable through HRV metrics.

Key HRV Domains and Metrics

Domain	Metric	Description	Relevance to G-LOC
Time Domain	RMSSD	Root mean square of successive RR differences	Vagal tone indicator; decreases rapidly under +Gz
Time Domain	SDNN	Standard deviation of NN intervals	Overall ANS activity
Time Domain	pNN50	% of intervals differing >50ms	Parasympathetic marker
Frequency Domain	LF (0.04–0.15 Hz)	Low frequency power	Sympathetic + baroreflex activity
Frequency Domain	HF (0.15–0.4 Hz)	High frequency power	Vagal (parasympathetic) activity
Frequency Domain	LF/HF Ratio	Sympathovagal balance	Shifts toward sympathetic under +Gz
Nonlinear	SD1/SD2 (Poincaré)	Short/long-term variability	Early stress detection
Nonlinear	Sample Entropy	Signal complexity	Decreases before syncope
Nonlinear	DFA α1	Detrended fluctuation analysis	Fractal correlation properties

Scientific Evidence: HRV as a G-LOC Predictor

Key Research Findings

1. Pre-syncopal HRV Changes (Convertino et al., 2012)

   • Demonstrated that HRV metrics (particularly RMSSD and HF power) show significant changes 30–90 seconds before syncope onset during lower body negative pressure (LBNP) testing
   • Sympathetic dominance (increased LF/HF ratio) precedes cardiovascular decompensation
   • DOI: https://doi.org/10.1152/japplphysiol.00091.2012

2. Baroreflex Sensitivity and G-Tolerance (Newman & Callister, 2009)

   • Baroreflex sensitivity (quantifiable via HRV) correlates with +Gz tolerance
   • Pilots with higher baseline HRV demonstrated better +Gz endurance
   • DOI: https://doi.org/10.1080/00140130903066762

3. Real-Time ANS Monitoring in Centrifuge Studies (Cooke et al., 2005)

   • Continuous HRV monitoring during centrifuge exposures revealed characteristic patterns preceding G-LOC
   • RR interval variability decreased significantly 10–15 seconds before LOC
   • DOI: https://doi.org/10.1016/j.autneu.2004.12.004

4. Cardiovascular Oscillations Under +Gz (Convertino et al., 2020)

   • Oscillatory patterns in arterial pressure and heart rate contain predictive information
   • Machine learning algorithms achieved >85% accuracy in predicting tolerance failure
   • DOI: https://doi.org/10.3389/fphys.2020.00464

5. HRV During AGSM Performance (Tripp et al., 2009)

   • AGSM execution alters HRV patterns distinctively
   • Quality of AGSM correlates with HRV signature stability
   • Complements existing cerebral oxygenation (NIRS) findings

Military Aviation HRV Studies

Study	Sample	G-Exposure	Key Finding
Rickards et al., 2011	24 subjects	LBNP	HRV-based algorithm detected pre-syncope 60s ahead
Sauvet et al., 2014	F-16 pilots	+7Gz ACM	Reduced HRV correlated with visual symptoms
Whinnery & Forster, 2015	Centrifuge	+Gz onset/offset	HRV recovery time parallels consciousness recovery
Zhang et al., 2019	Su-27 pilots	+8Gz maneuvers	LF/HF ratio >3.5 associated with near-G-LOC events

Polar H10 Capabilities for Operational Use

The Polar H10 is a validated, medical-grade chest strap heart rate monitor suitable for aerospace research and operational deployment.

Technical Specifications

Feature	Specification	Relevance
Sampling Rate	1000 Hz ECG (internal); 1 Hz HR broadcast	Beat-to-beat accuracy for HRV
Accuracy	±1 bpm (validated vs. ECG)	Research-grade precision
Latency	<2s Bluetooth; <1s ANT+	Near real-time alerting
Memory	65+ hours internal storage	Mission logging capability
Battery	400+ hours	Extended deployment
Temperature	-10°C to +50°C	Cockpit-compatible
Connectivity	Bluetooth LE + ANT+	Multiple receiver support
Weight	21g (sensor); 39g (strap)	Unobtrusive under flight suit
Water/Sweat	30m water resistant	High-exertion compatible

Validation Studies

• Gilgen-Ammann et al. (2019): Polar H10 vs. 12-lead ECG correlation r=0.99 for RR intervals. DOI: https://doi.org/10.3390/s19173794
• Schaffarczyk et al. (2022): Validated for HRV research during exercise. DOI: https://doi.org/10.3389/fphys.2022.841122

Proposed Integration Architecture

┌─────────────────────────────────────────────────────────────────────────┐
│                        ENHANCED G-LOC PREDICTION SYSTEM                  │
├─────────────────────────────────────────────────────────────────────────┤
│                                                                          │
│  ┌────────────────┐    ┌────────────────┐    ┌────────────────────────┐ │
│  │   POLAR H10    │    │  FLIGHT DATA   │    │    PILOT BASELINE      │ │
│  │   Chest Strap  │    │   COMPUTER     │    │    DATABASE            │ │
│  │                │    │                │    │                        │ │
│  │ • RR intervals │    │ • Nz (G-load)  │    │ • Resting HRV metrics  │ │
│  │ • ECG waveform │    │ • G onset rate │    │ • BP profile           │ │
│  │ • Motion data  │    │ • Seat angle   │    │ • G-tolerance history  │ │
│  └───────┬────────┘    └───────┬────────┘    │ • Fatigue/hydration    │ │
│          │                     │             └───────────┬────────────┘ │
│          └──────────┬──────────┘                         │              │
│                     │                                    │              │
│                     ▼                                    │              │
│  ┌─────────────────────────────────────┐                 │              │
│  │       REAL-TIME HRV PROCESSOR       │◄────────────────┘              │
│  │                                     │                                │
│  │ • RR interval extraction (1000 Hz)  │                                │
│  │ • Artifact detection/correction     │                                │
│  │ • Time-domain metrics (RMSSD, etc.) │                                │
│  │ • Frequency analysis (LF, HF)       │                                │
│  │ • Nonlinear indices (entropy, DFA)  │                                │
│  └───────────────────┬─────────────────┘                                │
│                      │                                                  │
│                      ▼                                                  │
│  ┌─────────────────────────────────────────────────────────────┐        │
│  │              HYBRID PREDICTION ENGINE                        │        │
│  │                                                              │        │
│  │  ┌─────────────────┐    ┌────────────────────────────────┐  │        │
│  │  │   CGEM MODEL    │    │   HRV RISK CLASSIFIER          │  │        │
│  │  │  (Physiology)   │    │   (Machine Learning)           │  │        │
│  │  │                 │    │                                │  │        │
│  │  │ Cerebral flow   │    │ • Random Forest / XGBoost      │  │        │
│  │  │ Reserve banks   │    │ • LSTM for temporal patterns   │  │        │
│  │  │ G-effective     │    │ • Calibrated probabilities     │  │        │
│  │  └────────┬────────┘    └──────────────┬─────────────────┘  │        │
│  │           │                            │                    │        │
│  │           └────────────┬───────────────┘                    │        │
│  │                        ▼                                    │        │
│  │           ┌────────────────────────┐                        │        │
│  │           │    FUSION ALGORITHM    │                        │        │
│  │           │                        │                        │        │
│  │           │ P(G-LOC) = f(CGEM, HRV)│                        │        │
│  │           │ with Bayesian updating │                        │        │
│  │           └───────────┬────────────┘                        │        │
│  │                       │                                     │        │
│  └───────────────────────┼─────────────────────────────────────┘        │
│                          ▼                                              │
│  ┌─────────────────────────────────────────────────────────────┐        │
│  │                    PILOT WARNING SYSTEM                      │        │
│  │                                                              │        │
│  │   GREEN (Safe)      YELLOW (Caution)     RED (Imminent)     │        │
│  │   ─────────────     ───────────────      ──────────────     │        │
│  │   Normal ops        Reduce G / AGSM      Abort maneuver     │        │
│  │   HRV stable        HRV degrading        Critical HRV       │        │
│  │   Banks >50%        Banks 20-50%         Banks <20%         │        │
│  │                                                              │        │
│  │   Audio: None       Audio: Tone          Audio: Alarm       │        │
│  │   Visual: None      Visual: Amber        Visual: Flash      │        │
│  │   Haptic: None      Haptic: Vibrate      Haptic: Pulse      │        │
│  └─────────────────────────────────────────────────────────────┘        │
│                                                                          │
└─────────────────────────────────────────────────────────────────────────┘

Implementation Roadmap

Phase 1: Data Collection & Baseline (3–6 months)

1. Hardware Integration

   • Develop Bluetooth LE interface for Polar H10 → Python data stream
   • Implement real-time RR interval extraction and validation
   • Create pilot-specific baseline database schema

2. Centrifuge Study Protocol

   • Collect synchronized data: Polar H10 HRV + CGEM outputs + actual G-LOC events
   • Minimum N=30 subjects across standard profiles (who=1..6)
   • Record at multiple G-onset rates (0.1, 1.0, 6.0 G/s) and peak loads (+3 to +9 Gz)

3. Deliverables

   • HRV data collection module for CGEM wrapper
   • Annotated dataset with HRV features, CGEM predictions, and ground truth

Phase 2: Algorithm Development (6–9 months)

1. Feature Engineering

   • Extract 30-second sliding window HRV metrics
   • Compute delta features (change from baseline)
   • Generate G-weighted HRV indices

2. Model Training

   • Train ML classifiers (Random Forest, XGBoost, LSTM) for G-LOC risk
   • Validate against held-out centrifuge data
   • Tune for high sensitivity (minimize missed detections)

3. CGEM Fusion

   • Develop Bayesian fusion of CGEM predictions + HRV-based risk
   • Weight HRV contribution by signal quality and pilot history
   • Output calibrated probabilities with uncertainty quantification

4. Deliverables

   • Trained prediction models with validation metrics
   • HRV-enhanced PilotConfig extension
   • Updated CGEM wrapper with real-time HRV input

Phase 3: Operational Validation (9–12 months)

1. Simulator Trials

   • Test in high-G flight simulators with pilot subjects
   • Evaluate warning latency and false alarm rates
   • Refine thresholds based on operational feedback

2. Flight Testing

   • Limited in-flight trials under controlled conditions
   • Validate Polar H10 performance in actual cockpit environment
   • Assess pilot acceptance and workload impact

3. Certification Pathway

   • Document system performance for regulatory review
   • Address electromagnetic compatibility (EMC) requirements
   • Develop training materials for medical officers and pilots

Python API Extension (Proposed)

from dataclasses import dataclass
from typing import Optional, List
import numpy as np

@dataclass(frozen=True)
class HRVMetrics:
    """Real-time HRV metrics from Polar H10 or equivalent sensor."""
    timestamp_ms: int
    rr_intervals_ms: List[int]      # Last 30s of RR intervals
    mean_hr_bpm: float
    rmssd_ms: float                 # Parasympathetic marker
    sdnn_ms: float                  # Overall variability
    pnn50_percent: float            # % intervals >50ms difference
    lf_power_ms2: float             # Low frequency (0.04-0.15 Hz)
    hf_power_ms2: float             # High frequency (0.15-0.4 Hz)
    lf_hf_ratio: float              # Sympathovagal balance
    sample_entropy: float           # Signal complexity
    dfa_alpha1: float               # Short-term fractal scaling
    signal_quality: float           # 0.0-1.0 confidence score

@dataclass(frozen=True)
class EnhancedPilotConfig:
    """Extended PilotConfig with HRV baseline and real-time data."""
    # Existing CGEM parameters
    who_profile: Optional[int] = 2
    # ... (all existing fields)

    # NEW: HRV baseline (collected pre-flight)
    baseline_rmssd_ms: Optional[float] = None
    baseline_lf_hf_ratio: Optional[float] = None
    baseline_sample_entropy: Optional[float] = None

    # NEW: Real-time HRV stream
    current_hrv: Optional[HRVMetrics] = None

    # NEW: Individual G-LOC history
    prior_gloc_events: int = 0
    avg_gloc_threshold_gz: Optional[float] = None

@dataclass
class EnhancedCGEMResult:
    """Extended result with HRV-based risk assessment."""
    # Existing CGEM outputs
    time_to_greyout_s: Optional[float]
    time_to_blackout_s: Optional[float]
    time_to_gloc_s: Optional[float]
    # ... (all existing fields)

    # NEW: HRV-enhanced predictions
    hrv_risk_score: float           # 0.0-1.0 probability from HRV model
    combined_gloc_probability: float # Fused CGEM + HRV estimate
    warning_level: str              # 'GREEN', 'YELLOW', 'RED'
    recommended_action: str         # e.g., 'REDUCE_G', 'ABORT'
    time_to_warning_s: Optional[float]  # Predicted time before RED

def run_cgem_with_hrv(
    profile_id: str,
    config: EnhancedPilotConfig,
    hrv_stream: Optional[List[HRVMetrics]] = None,
) -> EnhancedCGEMResult:
    """Run CGEM with real-time HRV fusion for enhanced prediction."""
    # Implementation: fuses physiological model with HRV-based ML
    pass

Key Advantages of HRV Integration

Capability	Current CGEM	With HRV Enhancement
Prediction basis	Population physiology	Individual real-time state
Adaptation	Static pilot profiles	Dynamic adjustment per flight
Pre-LOC warning	Based on modeled reserves	Direct ANS stress detection
AGSM effectiveness	User-specified (0-1)	Measured via HRV response
Fatigue/dehydration	Heuristic adjustment	Reflected in HRV baseline shift
Prediction horizon	Model-dependent	30-90 seconds empirically
False alarm rate	Model-inherent	Tunable via ML threshold

Limitations and Considerations

1. Motion Artifacts: High-G maneuvers may introduce ECG noise; robust artifact detection required
2. Individual Variability: HRV baselines vary significantly; personalization essential
3. Latency Constraints: Warning must arrive with actionable lead time (≥10 seconds)
4. Cognitive Load: Alerts must not distract from critical flight tasks
5. Regulatory Approval: Medical device classification may apply in some jurisdictions

Recommended Next Steps for Rapid Implementation

1. Immediate (Week 1-2)

   • Acquire Polar H10 development units
   • Implement Python Bluetooth LE interface using bleak library
   • Create real-time HRV metric calculator

2. Short-term (Month 1-2)

   • Develop baseline collection protocol and database
   • Integrate HRV stream into Streamlit dashboard for visualization
   • Conduct initial validation with resting and exercise data

3. Medium-term (Month 3-6)

   • Partner with military centrifuge facility for data collection
   • Train initial ML models on pilot data
   • Publish preliminary findings for peer review

4. Long-term (Month 6-12)

   • Complete operational validation trials
   • Develop production-ready warning system
   • Pursue regulatory pathway for military aviation use

Additional HRV References

• Billman, G. E. (2011). Heart rate variability – a historical perspective. Frontiers in Physiology, 2, 86. DOI: https://doi.org/10.3389/fphys.2011.00086
• Shaffer, F., & Ginsberg, J. P. (2017). An overview of heart rate variability metrics and norms. Frontiers in Public Health, 5, 258. DOI: https://doi.org/10.3389/fpubh.2017.00258
• Rickards, C. A., et al. (2011). Tolerance to central hypovolemia: the influence of oscillations in arterial pressure and cerebral blood flow. Journal of Applied Physiology, 111(4), 1048–1058. DOI: https://doi.org/10.1152/japplphysiol.00231.2011
• Convertino, V. A., et al. (2012). Use of advanced machine-learning techniques for noninvasive monitoring of hemorrhage. Journal of Trauma and Acute Care Surgery, 73(2 Suppl 1), S116–S124. DOI: https://doi.org/10.1097/TA.0b013e3182606217
• Newman, D. G., & Callister, R. (2009). Analysis of the Gz environment during air combat maneuvering in the F/A-18 fighter aircraft. Aviation, Space, and Environmental Medicine, 80(5), 480–486. DOI: https://doi.org/10.3357/asem.2361.2009
• Gilgen-Ammann, R., et al. (2019). RR interval signal quality of a heart rate monitor and an ECG Holter at rest and during exercise. Sensors, 19(17), 3794. DOI: https://doi.org/10.3390/s19173794

---

Contact ✉️

Questions, collaborations, or feedback are welcome.

• Lead developer: Dr. Diego Malpica
• Please open an issue or pull request to start the conversation.