AI digital twin framework achieves 96.2% accuracy in aircraft fault diagnosis

Researchers have developed an intelligent fault diagnosis framework for general aviation aircraft that leverages multi-fidelity digital twins and failure mode and effects analysis (FMEA) knowledge. The system, detailed in a paper on arXiv, addresses challenges like scarce real fault data and weak fault signatures by integrating four modules: high-fidelity flight dynamics simulation using JSBSim's 6-DoF engine, an FMEA-driven fault injection engine modeling 19 fault types, a multi-fidelity residual feature extraction framework, and an LLM-enhanced interpretable report generator. The digital twin generates 23-channel engine health monitoring data via semi-empirical sensor synthesis equations.

The multi-fidelity residual computation framework includes two paths: a high-fidelity path using paired-mirror residuals with nominal mirror trajectories for clean fault deviation signals, and a low-fidelity path using a GRU surrogate model for online real-time residual computation. A 1D-CNN classifier performs end-to-end diagnosis across 20 fault classes. Experiments show the paired-mirror residual scheme achieves a Macro-F1 of 96.2%, while the GRU surrogate scheme provides 4.3x inference acceleration at only 0.6% performance cost. Analysis across 24 schemes reveals that residual feature quality contributes approximately 5x more to diagnostic performance than classifier architecture, establishing a "residual quality first" design principle.