虚假关系
方位(导航)
降级(电信)
计算机科学
阶段(地层学)
基质(化学分析)
模式识别(心理学)
序列(生物学)
贝叶斯概率
一致性(知识库)
数据挖掘
回归
残余物
人工智能
数学
统计
鉴定(生物学)
特征(语言学)
偏最小二乘回归
后验概率
算法
回归分析
区间(图论)
感知
贝叶斯推理
作者
Jiayang Zhao,Deqiang He,Zhenzhen Jin,Song Zhang,Yanjun Chen,Yang Fu
标识
DOI:10.1088/1361-6501/ae7c9d
摘要
Abstract Rolling bearings are key components in rotating machinery, and accurate remaining useful life (RUL) prediction is essential for condition-based and predictive maintenance. However, bearing degradation is typically nonstationary, with ambiguous health-stage boundaries and varying late-stage degradation rates, which limits the practical applicability of data-driven methods. To address these challenges, this paper develops a stage-perceived RUL prediction framework. A degradation-sensitive health indicator (HI) is first constructed through minimum-redundancy feature screening, attention-based fusion, and spurious fluctuation correction. Then, a Bayesian hidden semi-Markov stage perception model is proposed to represent bearing degradation as latent health-state segments with duration constraints. By constructing the observation sequence from the corrected HI, local increment, and fluctuation intensity, the model estimates posterior state probabilities to identify stable stage transitions. Furthermore, a multi-center locally consistent support matrix regression model is developed for degradation prediction. The degradation sequence is embedded into matrix samples and divided into multiple local degradation regions, where local support matrix regressors are dynamically integrated according to sample-center consistency to adapt to varying late-stage degradation rates. Experiments on public full-life bearing datasets demonstrate that the proposed method achieves more reliable stage identification and higher RUL prediction accuracy than representative comparison methods.
科研通智能强力驱动
Strongly Powered by AbleSci AI