ORAC-NT v5.x: Optimal and Stable FDIR Architecture for Autonomous Spacecraft and Critical Systems

Autonomous systems operating in safety-critical environments must maintain operational integrity despite faults, sensor degradation, and dynamic stress. Traditional FDIR architectures rely on rule-based logic or predefined fault trees, which scale poorly with system complexity. This work proposes a Control Lyapunov Function (CLF) formulation using the instability metric $V = T - Q \cdot D$, where $Q$ represents detection reliability, $D$ represents decision capability, and $T$ represents accumulated system stress. The metric provides both a health indicator and a stability guarantee: the system is stable if and only if $V < 0$ (capability exceeds stress). Recovery actions are derived from the control law $u^* = \arg\min_u V(x, u)$. Evaluation across 9,000 adversarial simulation missions demonstrates 100% detection rate and zero false alarms under Silent Drift, Byzantine, Cascading, and Final Boss fault scenarios. GravOpt integration achieves 35.6% energy reduction on ResNet18 under thermal stress (87°C) while maintaining full detection performance. Hardware validation on Arduino Uno Q with MPU-6050 IMU over 10,000 continuous steps (8.3 minutes at 20 Hz) confirms real-time viability, with mean detection latency of 3.2 ms and 95.7% recovery rate under mechanical shocks. These results establish a compact, theoretically-grounded framework for resilient autonomous systems. **Index Terms**—FDIR, CubeSat, fault detection, Control Lyapunov Function, CUSUM, Byzantine consensus, energy optimization, autonomous systems, hardware validation, stability guarantees --- ## I. INTRODUCTION Autonomous systems deployed in remote or safety-critical environments must detect faults and perform recovery without continuous human supervision. This requirement is critical in aerospace missions and deep-space exploration, where communication delays prevent real-time operator intervention. At Neptune, for example, a signal requires approximately 4 hours to reach Earth – making ground control of fault scenarios impossible. Current spacecraft FDIR mechanisms rely on rule-based diagnostics, threshold checks, and predefined recovery procedures. While robust, these approaches require extensive engineering effort to enumerate fault conditions. As system complexity grows, manual rule construction becomes increasingly difficult. The Voyager and New Horizons missions operate with fixed thresholds, while ESA JUICE has limited onboard autonomy but no environmental ML model. This paper introduces a Control Lyapunov Function (CLF) formulation using the instability metric $V = T - Q \cdot D$. Unlike traditional health metrics, this formulation provides both a scalar health indicator and a stability guarantee: the system is stable if and only if capability exceeds stress ($Q \cdot D > T$). Recovery actions are derived from the control law $u^* = \arg\min_u V(x, u)$, providing a mathematically rigorous foundation for autonomous fault recovery. --- ## II. RELATED WORK Traditional FDIR in spacecraft relies on rule-based diagnostics and threshold monitoring. Early autonomy research demonstrated onboard planning feasibility: the Remote Agent architecture at JPL [2] integrated planning, execution, and monitoring into a unified framework. CUSUM detectors [3] provide statistical change detection, while Byzantine fault tolerance [4] addresses coordinated sensor failures. Recent work on sparse training (RigL, SET) and hardware-aware quantization addresses model compression, but not runtime adaptation to environmental conditions. Control Lyapunov Functions have been extensively used in nonlinear control [5] but rarely applied to fault detection and recovery. This work bridges CLF theory with practical FDIR implementation. ---