92 / 2025-04-14 16:05:28

A Novel Condition Identification Method for Degraded Bearing

condition identification, deep autoencoder, classifier, uncertainty

Accurately identifying bearing conditions is a key technology for enhancing the mission reliability of aerospace equipment and reducing maintenance costs. Considering the multi-stage characteristics of bearing performance degradation, existing methods generally use clustering or change-point detection techniques for condition identification. However, a common challenge is that traditional identification methods are limited to offline results with point estimation and have weak online identification capabilities considering uncertainty. To address this, this paper proposes a novel bearing condition identification method. First, the Fuzzy C-Means (FCM) clustering algorithm is used to perform offline analysis of the root mean square (RMS) features of historical bearings to obtain period labels for training samples. Meanwhile, a deep autoencoder integrated with a softmax classifier is trained using the time domain and frequency domain features of historical bearings. After training, the method enables online identification by inputting the extracted features of on-site bearings, with the introduced softmax classifier simultaneously determining the probabilities of the bearing belonging to each stage. Finally, the effectiveness of the proposed method is verified using public bearing datasets. Experimental results show that the method exhibits strong online identification capability and can accurately recognize bearing conditions.