关键词: 稀疏自编码器特征融合, 双向长短期记忆网络预测模型, 滚动轴承, 反双曲特征指标, 频域谐波退化因子

Abstract: To address the demands for remaining life prediction of rolling bearings, a prediction model was proposed based on SAE and BiLSTM network. Taking the full-life vibration data of rolling bearings as research object, a degradation index set was constructed by developing a hyperbolic inverse transformation-based health indicator and a frequency-domain harmonic degradation factor. The SAE was employed for feature fusion to extract key features and eliminate redundant information. Meanwhile, the BiLSTM model was utilized to capture temporal dependencies and achieve full-cycle life prediction. Experimental results demonstrate that the proposed model outperforms support vector regression, extreme learning machines, and convolutional neural networks in terms of smaller prediction errors and stronger generalization capabilities.

Key words:  , sparse autoencoder(SAE) feature fusion, bidirectional long short-term memory(BiLSTM) network predictive model, rolling bearing, inverse hyperbolic characteristic index, frequency domain harmonic degradation factor