Sensor redundancy for supporting aero-engine condition monitoring and fault analysis

To address the challenges of measurement uncertainty and sensor fault analysis under complex aero-engine operating conditions, this paper proposes a novel sensor redundancy method based on causal inference. Designed for large-scale, complex flight data, the method employs the Hybrid Variational Stochastic Gradient Hamiltonian Monte Carlo (HVSGHMC) approach, which integrates Structural Causal Models (SCM) to accurately capture causal relationships and model uncertainties. HVSGHMC first uses Variational Inference (VI) to efficiently approximate the posterior distribution, providing a fast and smooth initialization. However, as real-world aero-engine data often involve high nonlinearity and noise, VI alone is insufficient to capture the full complexity of the target distributions. To address this limitation, HVSGHMC introduces Stochastic Gradient Hamiltonian Monte Carlo (SGHMC) to refine the posterior, enabling precise uncertainty modeling and fault analysis in complex environments. This progressive approach ensures superior performance in challenging scenarios. Simulation experiments using real flight data from turbofan and turboshaft engines validate the method’s scalability and accuracy. The results demonstrate that HVSGHMC reduces average modeling errors by 42.4% compared to the baseline deep neural network (DNN). These findings highlight HVSGHMC as a practical and effective solution for improving posterior estimation while providing interpretable support for condition monitoring and fault analysis in complex aero-engine systems.