基于RCMFFDE和SSA-RVM的旋转机械损伤检测模型  

Damage detection model of rotating machinery based on RCMFFDE and SSA-RVM

作　　者：王显彬[1] 孙阳 WANG Xianbin;SUN Yang(Institute of General Aviation Industry,Fujian Chuanzheng Communications College,Fuzhou 350007,China;School of Mechanical Engineering,Zhejiang University of Technology,Hangzhou 310023,China)

机构地区：[1]福建船政交通职业学院通用航空产业学院,福建福州350007 [2]浙江工业大学机械工程学院,浙江杭州310023

出　　处：《机电工程》2025年第3期510-519,共10页Journal of Mechanical & Electrical Engineering

基　　金：福建省引导性科技计划项目(2023H0027)。

摘　　要：针对旋转机械系统的振动信号具有明显的非线性,严重影响故障特征提取从而导致其识别精度不佳的问题,建立了一种基于精细复合多尺度分数波动散布熵(RCMFFDE)、t-分布随机邻域嵌入(t-SNE)和麻雀搜索算法优化相关向量机(SSA-RVM)的旋转机械损伤检测模型。首先,进行了基于RCMFFDE方法的特征提取,生成了特征样本,以定量反映旋转机械的不同损伤情况;然后,采用t-SNE方法,将原始高维故障特征映射至低维空间,获得了对故障更敏感的低维特征;最后,将敏感的低维故障特征向量输入至SSA-RVM多分类器中,进行了训练和测试,实现了旋转机械样本的故障识别目的;采用两种旋转机械数据集进行了实验,并从准确率、效率和抗噪性方面,将RCMFFDE-SSA-SVM方法与多种特征提取方法进行了对比。研究结果表明:RCMFFDE能用于有效提取旋转机械的故障特征,分别取得99.2%和100%的识别精度;而对敏感特征进行分类所获得的精度优于对原始特征进行分类的情形,前者比后者提高了4%;在模式识别中,SSA-RVM优于其他分类器;自制数据集的诊断精度达到了97%,特征提取的时间为16.05 s。Aiming at the problem that vibration signals of rotating machinery systems have obvious nonlinearity,it seriously affects fault feature extraction and leads to poor identification accuracy.A rotating machinery damage detection model based on refined composite multi-scale fractional fluctuation dispersion entropy(RCMFFDE),t-distributed stochastic neighbor embedding(t-SNE),and sparrow search algorithm optimized relevance vector machine(SSA-RVM)was established.Firstly,based on the RCMMFFDE method,feature extraction was performed to generate feature samples that quantitatively reflected different damage conditions of rotating machinery.Then,the original high-dimensional fault features were mapped to a low dimensional space,low dimensional features were obtained using the t-SNE method that were more sensitive to faults.Finally,sensitive low dimensional fault feature vectors were input into the SSA-RVM multi classifier for training and testing,achieving fault recognition of rotating machinery samples.Experiments were conducted using two rotating machinery datasets and compared with various feature extraction methods in terms of accuracy,efficiency,and noise resistance.The results show that the RCMFFDE can effectively extract the fault features of the rotating machinery,and the recognition accuracy is 99.2%and 100%respectively.The accuracy obtained by classifying sensitive features is better than that obtained by classifying original features,with an accuracy increase of 4%.In the aspect of pattern recognition,SSA-RVM is superior to other classifiers.The diagnostic accuracy of the self-made dataset reaches 97%,and the feature extraction time is 16.05 s.

关 键 词：非线性振动信号 特征提取时间 故障识别精度(诊断精度) 精细复合多尺度分数波动散布熵 t-分布随机邻域嵌入 麻雀搜索算法优化相关向量机 

分 类 号：TH133[机械工程—机械制造及自动化]