测控技术2025，Vol.44Issue(7)：49-57,9.DOI:10.19708/j.ckjs.2025.02.213
基于多尺度竞争注意力GRU的滚动轴承RUL预测
Rolling Bearing RUL Prediction Based on Multi-Scale Competitive Attention GRU
张泽中 1马静 1陈烁宇 1吕宏楠 2左强3
作者信息
- 1. 北京长城航空测控技术研究所有限公司,北京 101111||状态监测特种传感技术航空科技重点实验室,北京 101111
- 2. 新乡航空工业(集团)有限公司,河南新乡 453019
- 3. 中国人民解放军93160部队,北京 100166
- 折叠
摘要
Abstract
Remaining useful life(RUL)prediction of bearings is crucial to ensuring the safe and efficient oper-ation of mechanical components.The accuracy of bearing RUL prediction depends on the model's ability to ex-tract features from the operational data of bearing.Traditional recurrent neural networks(RNNs)do not balance the contribution of input multi-scale features to prediction before modeling,which greatly reduces the model's ability to capture key bearing operation data information,and the prediction performance needs to be improved.Against this backdrop,a multi-scale competitive attention network(MSCA-Net)is proposed to improve RUL prediction accuracy.MSCA-Net extracts both short-term and long-term features for each input attribute and e-valuates their contributions on multiple scales.When constructing RUL prediction models,it is possible to focus on important input scales and distinguish hidden features of different importance levels,and then gated recur-rent unit(GRU)network is used to complete RUL prediction.Comparative experiments between MSCA-Net and other advanced data-driven models(LSTM,TCN,CNN-BiLSTM-AM)are conducted by using the XJTU-SY bearing dataset.The results show that MSCA-Net achieves better performance than other methods.关键词
剩余使用寿命预测/滚动轴承/多尺度特征/竞争注意力机制/门控循环单元Key words
RUL prediction/rolling bearing/multi-scale features/competitive attention mechanism/GRU分类
信息技术与安全科学引用本文复制引用
张泽中,马静,陈烁宇,吕宏楠,左强..基于多尺度竞争注意力GRU的滚动轴承RUL预测[J].测控技术,2025,44(7):49-57,9.
