摘要
针对传统滚动轴承故障特征提取及识别高度依赖先验知识及专家经验,导致其故障诊断的人工成本高及分类精度不够高的问题,提出一种层叠P阶多项式主成分分析方法实现滚动轴承故障的精确诊断。提出一种可适用于处理线性不可分数据的P阶多项式主成分分析法从滚动轴承的振动信号中自动学习去相关的低维特征;构建了层叠P阶多项式主成分分析网络,从去相关的低维特征中进一步增强学习更具可分辨性的特征,并通过反向优化过程,确保学习的特征不失真;采用K最近邻分类器对学习到的特征矢量进行分类,实现故障模式的辨识。通过滚动轴承故障数据库上的诊断试验验证了该方法的可靠性和有效性。
The traditional rolling bearing fault feature extraction and recognition highly rely on priori knowledges and expert experiences,resulting in its high labor cost and not enough accurate classification.A method of stacked P-order polynomial principal component analysis(SPPCA)was proposed to realize the accurate diagnosis of rolling bearing faults.First,a P-order polynomial principal component analysis(PPCA),which is applicable to deal with linear inseparable data,was presented to automatically learn the uncorrelated low-dimensional features from the vibration signals of rolling bearings.Next,a SPPCA network was built to further learn more discriminative features,using the back-forward optimization to ensure that learnt features are not distorted.Then,a K nearest neighbor classifier was used to classify the learnt feature vectors to identify the fault model.The experimental results on the database of rolling bearings faults verified the reliability and validity of the proposed method.
作者
牟亮
王凯
李彦
於辉
MOU Liang;WANG Kai;LI Yan;YU Hui(School of Manufacturing Science and Engineering,Sichuan University,Chengdu 610065,China)
出处
《振动与冲击》
EI
CSCD
北大核心
2019年第2期25-32,共8页
Journal of Vibration and Shock
基金
国家自然科学基金(51435011)
国家自然科学基金青年科学基金(51505309)
四川省应用基础研究计划(2015JY0172)
