Note: ORAC-NT stands for Autonomous Recovery Controller — Neural Telemetry. This is unrelated to the ORAC (Oxygen Radical Absorbance Capacity) antioxidant measurement method.

ORAC-NT v5.5 — Autonomous Fault Detection and Recovery System

Autonomous Fault Detection, Isolation, and Recovery (FDIR) architecture for spacecraft and critical embedded systems.

---

Overview

ORAC-NT introduces a Lyapunov-guided autonomous recovery architecture driven by a scalar health metric:

W = Q · D − T

where:

• Q = subsystem quality score (1 − fault score)
• D = operational diversity (1 − workload ratio)
• T = thermal/stress factor

Recovery stability is guaranteed via the Lyapunov function:

V = (W* − W)²

The planner selects actions that monotonically decrease V, providing formal stability guarantees — not just heuristic recovery.

---

Key Results

Fault Detection (9,000 adversarial missions)

Test	Missions	Detection Rate	False Alarms	Avg Latency
Silent Drift	1,000	100%	0	30.4 steps
Byzantine	1,000	100%	0	24.8 steps
Cascading	2,000	100%	0	13.2 steps
Final Boss	5,000	100%	0	16.6 steps
Total	9,000	DR = 100%	FA = 0	—

Lyapunov Planner vs Random Baseline (1,000 missions)

Metric	Random Planner	Lyapunov Planner	Δ
Recovery Rate	57.3%	100%	+42.7 pp
Avg Steps to Recovery	4.68	2.08	2.3× faster

Recovery Benchmark (100 runs)

Strategy	Avg Steps	Improvement
Random	10.86	—
ORAC Lyapunov	2.30	~79%

Dual Sensor Byzantine Isolation (500 missions)

Metric	Result
Correct isolations	100%
False isolations	0

GravOpt Phase 1 — Energy Optimization

Metric	Result
Energy reduction	16.1% vs baseline
Detection Rate maintained	100%
Peak savings (SURVIVAL mode)	66.7%

Hardware Validation — Arduino UNO Q + MPU-6050

Metric	Value
Total steps	10,000 (8.3 min @ 20 Hz)
Recovery rate under shocks	95.7%
Mean detection latency	3.2 ms
False positives	0

---

Architecture

┌─────────────────────────────────────────────────────┐
│                   ORAC-NT v5.5                       │
│                                                      │
│  Sensors → Watchdog_v5 → OracController_v5           │
│               (CUSUM)      (W = Q·D − T)             │
│                                │                     │
│                         LyapunovPlanner              │
│                         V = (W* − W)²                │
│                         argmin_a V(W_predicted)      │
│                                │                     │
│                    NORMAL / SURVIVAL / RECOVERY       │
└─────────────────────────────────────────────────────┘

Watchdog_v5 — CUSUM-based fault detector with Byzantine majority voting (3 sensors). OracController_v5 — Computes W and switches operational mode. LyapunovPlanner — Greedy minimization of V; guarantees monotone recovery. DualSensorFusion — Isolates faulty sensor from two physical IMUs. GravOptController — Adaptive parameter freezing driven by W (16.1% energy savings). MicroSafe-RL — STM32F401 safety layer (~1.2 µs latency, 24 bytes RAM, MISRA-C compliant).

---

Hardware Validation (TRL 4)

Tested on Arduino UNO Q (Qualcomm QRB2210 + STM32U585) with MPU-6050 IMU:

Board:   Arduino UNO Q
Sensor:  ZX-MPU6050 (GY-521), I²C
Wiring:  VCC→3.3V  GND→GND  SDA→A4  SCL→A5

• Real gravity reading: gz ≈ 1.034g ✓
• Auto-calibration: 100-step per-axis baseline ✓
• Physical motion detection → SURVIVAL mode ✓
• Recovery to NORMAL after motion stops ✓

---

File Structure

orac_core.py              # Core — Watchdog_v5, OracController_v5, FaultGenerator
orac_final_proof.py       # Full benchmark — 9,000 missions, DR 100%, FA 0
orac_lyapunov_dual.py     # Lyapunov planner + dual sensor Byzantine test
orac_gravopt_phase1.py    # GravOpt Phase 1 — 16.1% energy reduction
orac_hardware.py          # Hardware — Arduino UNO Q + MPU-6050 (single sensor)
orac_hardware_dual.py     # Hardware — dual MPU-6050 Byzantine isolation
orac_benchmark.py         # Recovery benchmark — ~79% improvement vs random
ORAC_CubeSat.py           # CubeSat orbital simulation
orac_visual_simulator.py  # Visual simulator

---

Quick Start

pip install numpy matplotlib pyserial

# Run full benchmark (simulation)
python orac_final_proof.py

# Run Lyapunov planner + dual sensor test
python orac_lyapunov_dual.py

# Run recovery benchmark
python orac_benchmark.py

# Run hardware test (requires Arduino on COM4)
python orac_hardware.py

# Run dual sensor hardware test (requires 2x MPU-6050)
python orac_hardware_dual.py

---

Scientific Contributions

C1 — Health Metric Scalar vitality function W = Q·D − T for real-time spacecraft state estimation. Derived from the principle of useful work: capacity minus dissipation.

C2 — Lyapunov-Guided Recovery Architecture Planner selects actions with ΔV ≤ 0, providing formal stability guarantees. Integrates: health metric + stochastic fault detection + real-time planning.

C3 — Empirical Validation 9,000 adversarial missions, DR 100%, FA 0. Hardware-validated on real IMU (TRL 4), 3.2 ms detection latency.

---

Related Work

• FAAST — Facilitating Autonomy in Astrodynamics (CORDIS 101063274)
• LUMIO — ESA 12-U CubeSat lunar mission
• NASA Remote Agent — autonomous spacecraft control
• Google Project Suncatcher — orbital AI compute (TPU bit-flip fault context)

---

Citation

@misc{kretski2026oracnt,
  author    = {Kretski, Dimitar},
  title     = {ORAC-NT v5.x: Optimal and Stable FDIR Architecture for
               Autonomous Spacecraft and Critical Systems},
  year      = {2026},
  doi       = {10.5281/zenodo.20248197},
  publisher = {Zenodo}
}

---

Author

Dimitar Kretski — Independent researcher, Varna, Bulgaria GitHub: github.com/Kretski Zenodo: 10.5281/zenodo.20248197 Email: kretski1@gmail.com

---

ORAC-NT is a simulation and research prototype (TRL 4). Not yet qualified for flight use. Patent pending — Bulgarian Patent Office, filed 05.12.2025 (SynergyFF).