【作者】 王凯；

【导师】 纪国宜；

【作者基本信息】 南京航空航天大学 ， 工程力学， 2020， 硕士

【摘要】 机械故障诊断中,由于各种原因,采集的信号含有强噪声,导致故障信号特征微弱。以滚动轴承为例,采集到的轴承早期故障信号就是一种典型的微弱故障特征信号。所以寻找有效的方法提取并识别滚动轴承早期故障信号特征是一项重要的工作。在本文中,通过把提取信号特征的投影神经网络方法、稀疏分解方法和基于学习字典的训练方法三种方法互相结合,建立了一种轴承信号特征提取的混合类方法。具体过程如下:首先,把采集的正常轴承信号作为样本信号,从样本信号中选取两段信号,其中一段信号用来构成一个初始字典,另一段作为在初始字典上待分解的信号,通过投影神经网络优化算法求得样本信号中一段信号在初始字典上的稀疏系数;其次,把求得的样本信号的稀疏系数固定,采用交替方向乘子法训练初始字典,完成对字典的学习;最后通过运用稀疏分解方法中的稀疏度自适应匹配追踪算法求取故障信号在字典上的稀疏重构信号,再利用频谱图识别轴承故障频率,完成诊断。提出的轴承信号特征提取的混合类方法分别应用于仿真轴承信号和实际故障信号的诊断,其中实际故障信号通过搭建一个小型的航空发动机转子模型试验台来采集。通过把该方法分别运用于仿真信号和实际信号两种情形,数值实验说明运用提出的轴承信号特征提取的混合类方法可以得到轴承故障信号特征,从而完成轴承故障诊断。更多还原

【Abstract】 In the mechanical fault diagnosis,due to various reasons,the collected signal contains strong noise,resulting in weak fault signal characteristics.Taking rolling bearing as an example,the collected early fault signal of bearing is a typical weak fault characteristic signal.So it is an important work to find an effective method to extract and identify the early fault signal characteristics of rolling bearing.In this paper,a hybrid method of bearing signal feature extraction is established by combining three methods of signal feature extraction including projection neural network method,sparse decomposition method and dictionary learning method.The specific process is as follows.Firstly,we take the collected normal bearing signal as the sample signal,then,we select two signals from the sample signal,one of which is used to form an initial dictionary,the other is used as the signal to be decomposed on the initial dictionary,and obtain the sparse coefficient of a signal in the sample signal on the initial dictionary through the projection neural network optimization algorithm;secondly,we fix the sparse coefficient of the sample signal which has obtained from the first step,and then,we train the initial dictionary by the alternating multiplier method to complete the dictionary learning;finally,the sparse reconstruction signal of the fault signal in the dictionary is obtained by using the sparsity adaptive matching pursuit algorithm,and the bearing fault frequency is further identified by using the spectrum diagram to complete the diagnosis of the signal fault.The hybrid method of bearing signal feature extraction is applied to the diagnosis of simulated bearing signal and actual fault signal,respectively.The actual fault signal is collected by building a small aeroengine rotor model test-bed.By applying presented method to the simulation signal and the actual signal,the numerical experiments show that the hybrid method of bearing signal feature extraction can get the bearing fault signal features,thus completing the bearing fault diagnosis.更多还原