基于RCMFME和AO-ELM的齿轮箱损伤识别策略

Gearbox damage identification strategy based on RCMFME and AO-ELM

针对模糊熵只考虑信号的局部特征而忽略信号的全局特征,导致齿轮箱故障识别的准确率不佳的问题,提出了一种基于精细复合多尺度模糊测度熵(RCMFME)、天鹰优化器(AO)优化极限学习机(ELM)的齿轮箱故障诊断方法。首先,在精细复合多尺度模糊熵的基础上,对矢量的构造方式进行了改进,提出了能够同时考虑时间序列局部特征和全局特征的RCMFME方法;随后,利用RCMFME指标提取了齿轮箱振动信号的熵值,组建了故障特征向量;接着,利用AO算法对极限学习机的参数进行了自适应搜索,生成了参数最优的多类别分类器;最后,将训练样本的故障特征向量输入至AO-ELM分类模型中进行了模型训练,以构造性能最优的分类器,并实现了对齿轮箱测试样本的故障识别目的;利用两种齿轮箱振动数据集进行了实验,在识别准确率和识别稳定性方面,与相关的特征提取方法进行了对比。研究结果表明:采用基于RCMFME和AO-ELM的故障诊断方法能够分别取得100%和98%的分类准确率,平均识别准确率分别达到了100%和98%,优于精细复合多尺度全局模糊熵(RCMGFE)、精细复合多尺度模糊熵(RCMFE)、精细复合多尺度样本熵(RCMSE)。该方法具有显著的应用潜力。

Aiming at the defect that the fuzzy entropy only considered the local feature of the signal and ignored the global feature of the signal,which led to the poor accuracy of gearbox fault identification,a gearbox fault diagnosis method based on the refined composite multiscale fuzzy measure entropy(RCMFME)and the aquila optimizer(AO)optimized extreme learning machine(ELM)was proposed.Firstly,based on the refined composite multiscale fuzzy entropy,a new RCMFME method was proposed to consider both local and global features of time series by improving the construction of vectors.Subsequently,RCMFME index was used to extract the entropy value of the gearbox vibration signal and construct a fault feature vector.Then,the AO algorithm was used to adaptively search for the parameters of the extreme learning machine,generating a multi class classifier with the best parameters.Finally,the fault feature vectors of the training samples were input into the AO-ELM classification model,and the model was trained to construct the best performing classifier,achieving the goal of fault recognition for gearbox test samples.Experiments were conducted using two types of gearbox vibration datasets,and the recognition accuracy and stability were compared with relevant feature extraction methods.The research results show that the fault diagnosis methods based on RCMFME and AO-ELM can respectively achieve 100%and 98%classification accuracy,and the average recognition accuracy respectively reaches 100%and 98%,which is superior to refined composite multiscale global fuzzy entropy(RCMGFE),refined composite multiscale fuzzy entropy(RCMFE),refined composite multiscale sample entropy(RCMSE),and it has significant application potential.