A Data-Driven Prototype-Constrained Radial Response Network for Robust Aero-Engine Fault Diagnosis
Abstract
Abstract:Reliable fault diagnosis of aero-engine bearings is crucial for ensuring flight safety. However, the spectral distributions of different operating states are often distorted by noise and operating-condition variations, making it difficult for existing intelligent diagnosis methods to accurately characterize their intrinsic distribution structures. To address this issue, this paper proposes a Prototype-Constrained Radial Response Network (PCRRN) for aero-engine bearing fault diagnosis based on vibration spectrum analysis. First, representative prototypes are learned to characterize the intrinsic distribution structures of different operating states in a transformed feature space. Subsequently, a similarity-driven radial response modeling is established to quantify the relationships between testing spectra and the learned prototypes, thereby converting the original spectral features into compact prototype-response representations. Finally, a regularized response-space decision model is constructed to accomplish multi-state fault classification through an analytical least-squares optimization. Experimental validation is conducted on the HIT aero-engine bearing benchmark dataset under five signal-to-noise ratio conditions ranging from 0 dB to −8 dB. Comparative results demonstrate that the proposed PCRRN consistently outperforms various intelligent diagnostic methods under different noise conditions. The experimental results verify that the proposed prototype-constrained similarity learning framework effectively improves the robustness, discriminative capability, and computational efficiency of fault diagnosis under noisy operating environments, demonstrating its potential for real-time condition monitoring, intelligent maintenance decision-making, and safety assurance of aero-engine rotating machinery in practical engineering applications. Conflict of Interest Statement The authors declare no conflicts of interest.