Answers are grounded in this paper and cite their sources.
This paper presents PHI-Twin, an open cyber-physical pipeline and 11-class fault taxonomy for real-time Dornier 228 landing gear health monitoring, combining a MATLAB Simscape Multibody plant with a dual-input long short-term memory (LSTM) classifier behind a confidence-weighted fusion layer ([Formula: see text]) that falls back to a plant-reference class when AI uncertainty exceeds the threshold. The plant generates 9500 physics-based landing cycles across 11 Air Transport Association (ATA)-32-aligned fault classes. The LSTM’s 98.5% accuracy reflects taxonomy-reconstruction fidelity under matched synthetic assumptions, not field diagnostic performance; the core contribution is the integrated pipeline itself—real-time FlightGear telemetry, autonomous Simscape synchronization, PostgreSQL provenance logging, and PyVista/Grafana visualization. CS-23.473 sink-rate and load-factor thresholds are evaluated as deterministic checks independent of AI confidence, a baseline guarantee rather than the architecture’s novel claim. Calibration analysis yields an expected calibration error of 0.017 (0.0045 after temperature scaling), with a high-confidence fault miss rate of 0.50% at [Formula: see text]. Five simulated missions validate end-to-end fallback operation and cross-simulator telemetry integration, with 47 ms inference-to-display latency, at under $6700—a 15- to 45-fold reduction against commercial health and usage monitoring systems retrofit costs. All metrics derive from synthetic physics-model-generated conditions; the architecture and open-source pipeline are presented as a certification-aware research template, with independent hardware validation and field data collection identified as the necessary next steps toward a genuine diagnostic performance claim.
No takes yet. Share an insight, caveat, or question.
Sani et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: