基于EMD和Multi-fractal spectrum的BP水机故障诊断

为了准确判断水轮机组的故障,提高水轮机组诊断的精确性,建立了EMD-Multi-fractal spectrum和改进BP神经网络相结合的机组振动故障诊断模型.选取水轮发电机组不同工况下的轴系正常、轴承油膜涡动、转子部件不平衡、转子不对中等状态,采集各状态下的振动信号.经过经验模态分解得到振动信号波各种故障信号的EMD分量,根据信号波形趋势图由EMD系数提取出波形样本,再由多重分形谱算法提取波形样本的特征值alpha(q), f(q),将该特征向量作为BP神经网络的输入进行分类识别.将训练好的神经网络应用于全部样本,得到测试正确率为100%.该模型用波形提取信号特征代替了传统的频谱特性,并结合先进的多重分形谱进行诊断识别,为水轮发电机组故障诊断提供了一种新的思路.应用信号采集于水电厂运行的水轮机,根据诊断的结果对轴系各个部件进行局部校正,通过检测发现振动和摆度都大大减弱.该方法提高了检测精度,增强了人机交互性,具有重要的理论意义和实用价值.

An investigation into fault diagnosis of hydro-turbine unit based on EMD multifractal spectrum

To diagnose faults in of a hydro-turbine unit accurately and precisely, a fault diagnosis model for vibration signal of hydropower units is built based on the EMD, multi-fractal spectrum and improved BP neural network in this paper. A series of vibration signals under various conditions, such as normal rotor system, oil film whirl in bearings, rotor imbalance, and rotor misalignment etc. are acquired from a hydro-turbine unit. At first, the EMD components of these vibration signals are obtained through empirical mode decomposition. Then the waveform samples are extracted by using EMD coefficients according to the signal waveform tendency curves. Thirdly, the eigenvalues alpha(q)and f(q), are extracted from the waveform samples by means of multifractal spectrum algorithm. Finally, the eigenvectors are input into a BP network for classification and recognition. The trained neural network is applied to all the samples and the test accuracy is 100%. The results show that the multi-fractal spectrum not only is feasible for fault diagnosis of hydropower unit but also can improve the precision of diagnosis and enhance human-computer interaction.