基于振动特性的转轴裂纹识别研究

发布时间：2018-03-21 03:38

本文选题：裂纹识别　切入点：悬臂轴　出处：《西南交通大学》2017年硕士论文　论文类型：学位论文

【摘要】：旋转机械是现代社会中最重要的机械设备之一,它在各个工程行业中得到了广泛的应用。转子则是整个旋转机械中最为核心的部件,转子的正常与否直接决定整个旋转机械能否正常运转。转子裂纹故障是旋转机械最常见故障形式,对转子进行裂纹识别并判断裂纹的位置和深度,从而及时采取相应的措施实现预防性维护,对于保证设备安全高效的运行具有重要的社会意义和经济价值。本文基于断裂力学理论推导和有限元法,对转轴进行了裂纹识别研究。在对转轴的裂纹识别研究中,首先对悬臂裂纹轴和转子裂纹轴进行了振动测试实验研究。通过实验研究裂纹对转轴振动特性参数的影响,并测得含有各个裂纹工况的裂纹转轴水平方向和竖直方向的固有频率。通过对测得不同裂纹工况的悬臂轴两个方向上的固有频率值分析,得到由于裂纹引起裂纹轴固有频率具有不对称性特征结论。以裂纹悬臂轴为基础研究对象,通过参数化建模循环仿真出大量含不同裂纹位置和深度悬臂轴的模态参数,并将所有裂纹参数与模态参数作为样本形成样本数据库。然后在MATLAB中设计搭建裂纹识别神经网络,从实验研究中得到的裂纹轴固有频率具有不对称性特征结论,因此选取裂纹轴水平方向和竖直方向的固有频率同时作为神经网络的输入。通过从样本数据库中选取样本训练神经网络,并用实验数据修正及验证裂纹识别神经网络的准确性,最后得到基于振动特性的悬臂轴裂纹识别方法。并与等高线法作对比,本文方法在裂纹深度识别精度上平均高1.76%,在裂纹位置识别精度上平均高2.61%,充分说明利用具有不对称振动特性参数对裂纹识别的优越性。最后在悬臂轴裂纹识别研究的基础上,通过对转子系统模型精准化有限元建模及对神经网络的各项参数的调节,得到一套完善的转子裂纹识别方法。在转子系统模型精准化建模时,通过等效刚度弹簧单元模拟其轴承约束,并通过实验数据修正转子裂纹轴有限元模型,最终得到精准的转子裂纹系统模型。在神经网络训练时,通过对神经网络的隐含层节点数、权值、阈值、学习目标和学习速度等参数的调节,最后优化得到能够适用于转子裂纹识别研究的神经网络。本文基于具有不对称振动特性参数能实现对悬臂轴和转子系统的裂纹识别,并具有较高的精度,而且可以通过改变有限元模型及调节神经网络后能够适用多种裂纹识别研究。
[Abstract]:Rotating machinery is one of the most important mechanical equipment in modern society, and it has been widely used in various engineering industries. The rotor is the most important part in the whole rotating machinery. Whether the rotor is normal or not directly determines whether the whole rotating machinery can run normally. The rotor crack fault is the most common fault form of rotating machinery. The rotor crack is identified and the position and depth of the crack are judged. Therefore, taking corresponding measures to realize preventive maintenance in time is of great social significance and economic value to ensure the safe and efficient operation of the equipment. This paper is based on the theory of fracture mechanics and finite element method. In the research of crack identification of rotating shaft, the vibration test of cantilever crack shaft and rotor crack shaft is carried out. The influence of crack on the vibration characteristic parameter of rotating shaft is studied by experiment. The natural frequencies of the horizontal and vertical direction of the crack shaft are measured, and the natural frequencies of the cantilever shaft under different crack conditions are analyzed. It is concluded that the natural frequency of the crack shaft is asymmetric due to the crack. Taking the crack cantilever shaft as the research object, a large number of modal parameters with different crack positions and depths are simulated by parameterized modeling cycle. All the crack parameters and modal parameters are taken as samples to form a sample database, and then a crack identification neural network is designed in MATLAB. The results obtained from the experimental study show that the natural frequency of the crack axis is asymmetric. Therefore, the natural frequency of the horizontal and vertical direction of the crack axis is chosen as the input of the neural network. The neural network is trained by selecting the samples from the sample database, and the accuracy of the neural network is corrected and verified by the experimental data. Finally, the method of crack identification for cantilever shaft based on vibration characteristics is obtained, and compared with the contour method. In this paper, the average accuracy of crack depth identification is 1.76 and that of crack location is 2.61, which fully illustrates the superiority of using the parameters with asymmetric vibration characteristics to identify crack. Finally, on the basis of the research on crack identification of cantilever shaft, A set of perfect rotor crack identification method is obtained by accurate finite element modeling of rotor system model and adjustment of the parameters of neural network. The bearing constraint is simulated by equivalent stiffness spring element, and the finite element model of rotor crack shaft is modified by experimental data. Finally, a precise rotor crack system model is obtained. By adjusting the number of hidden layer nodes, weights, thresholds, learning goals and learning speed of the neural network, Finally, a neural network which can be applied to the research of rotor crack identification is obtained. Based on the parameters with asymmetric vibration characteristics, the crack identification of cantilever shaft and rotor system can be realized in this paper, and the accuracy is high. Moreover, the finite element model can be changed and the neural network can be adjusted to study the crack identification.
【学位授予单位】：西南交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TH17

【相似文献】