基于ARMA及神经网络的汽轮机振动故障诊断研究

根据Bently实验台所采集的碰摩、松动、不对中、不平衡4种典型的汽轮机转子振动故障水平方向与垂直方向的数据所建立的汽轮机转子振动故障序列自回归滑移平均(ARMA)模型,由ARMA模型参数计算自谱函数值,建立汽轮机转子振动故障时间序列的自谱函数图谱。对不同类故障所建立ARMA模型的自谱函数图谱分析表明:故障征兆信息较明显,有较好的故障区分度。另外由于ARMA模型的特征向量浓缩了原时间序列信号的全部信息,对ARMA模型的特征向量参数利用多节点输入双隐层BP神经网络完成p维欧氏空间到二维欧氏空间的非线性映射,对汽轮机转子振动故障状态进行诊断。诊断结果表明:对应故障类型的ARMA模型样本通过训练后的神经网络在二维欧氏空间中能较好地对故障进行分类,同类故障的检验样本与目标函数值在欧氏空间具有最小距离,表明基于ARMA模型的二维欧氏空间双隐层神经网络故障诊断方法有较高的故障辨识能力。

A Study of Turbine Vibration-fault Diagnosis Based on an ARMA and a Neural Network

Based on an ARMA(auto-regression moving average) model for turbine rotor vibration-fault series,the authors have calculated the self-spectral functional values through the use of ARMA model parameters and established a self-spectral function atlas for turbine rotor vibration-fault time series.The model has been established by using the data of four typical turbine rotor vibration faults all acquired on a Bently test rig in both horizontal and vertical directions,namely,rubbing,loosening,misalignment and unbalance.An analysis of the above atlas of the ARMA model set up for different kinds of faults shows that the information featuring fault symptoms is relatively clear,displaying a comparatively good division between various fault zones.Moreover,as the eigenvectors of the ARMA model have concentrated all the information of the original time series signals,a non-linear mapping for the eigenvector parameters of the ARMA model from a p-dimensional Euclidean space to a two-dimensional one has been performed by using a multi-node input dual hidden-layer BP neutral network in order to conduct a diagnosis of the status of a turbine rotor vibration fault.The diagnosis result indicates that the specimens of the ARMA model for a corresponding type of faults can relatively well classify the faults in a two-dimensional Euclidean space through the use of a neutral network which has undergone a training.The inspection specimen for the same type of faults have kept a minimal distance from the target functional value in a Euclidean space.The foregoing shows that the ARMA model-based fault diagnostic method of a two-dimensional Euclidean space and dual hidden-layer neutral network has a comparatively high ability to discriminate faults.