Intelligent sensor-based multi-fault diagnosis of aero-engines using scattering characteristics

Abstract

To address the problem of insufficient real-time fault diagnosis accuracy in distributed intelligent sensors for aero-engine systems under multiple coexisting fault modes and limited computational resources, an intelligent sensor fault diagnosis method based on local data scattering feature optimization is proposed. First, a wavelet scattering network is employed to extract translation-invariant deep features from sensor signals, enabling efficient characterization of their time–frequency structures and energy distributions in resource-constrained environments. Subsequently, Principal Component Analysis (PCA) is applied to reduce the dimensionality of the high-dimensional scattering features, thereby removing redundant information and enhancing feature discriminability, which effectively improves diagnosis accuracy. Finally, a Support Vector Machine (SVM) multi-classification model is constructed based on the reduced features to achieve rapid fault-type identification, satisfying the real-time diagnostic requirements. Experimental results demonstrate that for typical sensor faults, including abrupt changes, drifts, offsets, and periodic disturbances, the proposed method achieves an average diagnosis accuracy of 98.2%, representing a 6.5% improvement compared with direct use of the original features. Moreover, the average diagnosis time per sample is only 32 ms, meeting the demands of high accuracy, low computational complexity, and real-time performance in aero-engine sensor applications.