Abstract

The fault diagnosis of general aviation aircraft is a complex challenge due to several factors, including the limited availability of real fault data, the variety of fault types that can occur, and the often weak signatures that these faults produce. This paper introduces an innovative fault diagnosis framework that leverages a multi-fidelity digital twin approach. The framework consists of four key modules: high-fidelity flight dynamics simulation, fault injection driven by Failure Mode and Effects Analysis (FMEA), multi-fidelity residual feature extraction, and a large language model (LLM) that enhances the generation of interpretable diagnostic reports.

Core Methodology

The digital twin is constructed using the JSBSim six-degree-of-freedom (6-DoF) flight dynamics engine, which simulates the aircraft’s behavior under various conditions. This simulation generates 23 channels of engine health monitoring data through semi-empirical sensor synthesis equations, which are mathematical models that relate sensor outputs to the physical state of the aircraft.

To address the challenge of fault diagnosis, the authors developed a three-layer fault injection engine based on FMEA. This engine models the physical causal propagation of 19 different engine fault types, allowing for a systematic approach to understanding how faults might develop and affect aircraft performance.

Furthermore, the paper proposes a multi-fidelity residual computation framework that includes paired-mirror residuals and a Gated Recurrent Unit (GRU) surrogate prediction model. The high-fidelity path utilizes nominal mirror trajectories with identical initial conditions to extract clean fault deviation signals, while the low-fidelity path enables online real-time residual computation through the GRU surrogate model, which predicts residuals over multiple steps.

For the classification of faults, a 1D Convolutional Neural Network (CNN) is employed to perform end-to-end diagnosis across 20 different fault classes. The integration of an LLM diagnostic report engine, enhanced with FMEA knowledge, allows for the fusion of classification results, residual evidence, and domain-specific causal knowledge. This results in the generation of natural language reports that are interpretable and useful for engineers and technicians.

Experimental results demonstrate that the paired-mirror residual scheme achieves a Macro-F1 score of 96.2% on the 20-class fault diagnosis task. Additionally, the GRU surrogate scheme provides a significant inference acceleration of 4.3 times while incurring only a 0.6% performance cost. A comparative analysis across 24 different schemes reveals that the quality of residual features has a much greater impact on diagnostic performance—approximately five times more—than the architecture of the classifier itself. This finding establishes the principle of prioritizing residual quality in the design of fault diagnosis systems.

Why this matters for the future

This research is significant for the future of aviation safety and maintenance, particularly in the realm of general aviation, where resources and data are often limited. By utilizing a multi-fidelity digital twin approach, the framework not only enhances the accuracy of fault diagnosis but also provides a scalable solution that can adapt to various aircraft models and fault types. The integration of advanced machine learning techniques, such as CNNs and GRUs, with traditional engineering methodologies like FMEA, represents a crucial step towards more intelligent and automated maintenance systems.

Moreover, the ability to generate interpretable diagnostic reports using LLMs can bridge the gap between complex data analysis and practical application in the field. This means that maintenance personnel can make informed decisions based on clear, understandable reports, ultimately leading to improved aircraft safety and reduced downtime.

Conclusion

In conclusion, the proposed intelligent fault diagnosis method for general aviation aircraft represents a significant advancement in the field of aerospace engineering. By addressing the challenges of fault diagnosis through a multi-fidelity digital twin framework and integrating FMEA knowledge, this research offers a robust solution that enhances diagnostic accuracy and efficiency. The findings underscore the importance of residual quality in diagnostic systems and pave the way for future innovations in aircraft maintenance and safety protocols. As the aviation industry continues to evolve, such intelligent systems will be essential in ensuring the reliability and safety of aircraft operations.