基于敏感瞬态冲击特征提取与分层极限学习机的行星齿轮箱故障诊断研究

FAULT DIAGNOSIS OF PLANETARY GEAR BOX BASED ON SENSITIVE TRANSIERT IMPACT FEATURE EXTRACTION AND HIERARCHICAL EXTREME LEARNING MACHINE

在行星齿轮箱故障智能诊断中,针对其振动信号特征"难提取"、构建的特征向量集"质量差"以及基于极限学习机的故障分类模型"精度低"的问题,提出一种如何捕获其振动信号中敏感瞬态冲击特征并据此构建高维特征向量集与提升极限学习机故障分类精度的行星齿轮箱太阳轮的状态辨识方法。首先,将所测取振动信号分别经快速峭度图求解和变分模态分解,筛选出与其最大峭度值对应中心频率fω相匹配的若干个本征模函数,然后,求其改进多尺度排列熵值来构建高维特征向量集;其次,利用去噪自动编码器使极限学习机隐含层节点的输入权值和阈值满足正交条件实现其隐含层的分层;最后,将上述特征向量集作为分层核极限学习机的输入,通过训练建立行星齿轮箱太阳轮的故障分类模型。结果表明,所提方法实现了太阳轮振动信号中敏感瞬态冲击特征的有效提取及其特征向量集的高质量构建,同时也提高智能诊断模型的分类精度。

In the intelligent diagnosis of planetary gearbox faults,the issue of"difficult extraction"of the vibration signal characteristics,the"quality difference"of the constructed eigenvector set and the"low precision"of the fault classification model based on the extreme learning machine.Put forward a state identification method for a planetary gearbox solar wheel for how to capture sensitive transient impact feature in the vibration signal and construct a high-dimensional eigenvector set and improve the fault classification accuracy of the extreme learning machine.Firstly,the vibration signals are respectively solved by fast kurtosis diagram and variational mode decomposition and several intrinsic mode function matching the center frequency fωcorresponding to the maximum kurtosis value are selected,find the value of improve multi-scale permutation entropy to construct the highdimensional eigenvector set.Secondly,the de-noising automatic encoder is used to make the input weight and threshold of the implicit learning node of the extreme learning machine satisfy the orthogonal condition to realize the layering of its hidden layers.Finally,the above eigenvector set is used as the input of the hierarchical extreme learning machine,and the fault classification model of the planetary gearbox solar wheel is established through training.The results show that the proposed method achieves the effective extraction of sensitive transient impact feature in the vibration signal of the solar wheel and the high quality construction of the eigenvector set,and also improves the classification accuracy of the intelligent diagnosis model.