基于递归神经网络的旋转机械故障诊断方法

为实现对旋转机械的在线故障诊断,对10类故障情况下的振动信号进行频谱分析。发现旋转机械振动信号的频谱中含有丰富的故障信息,以此为故障特征向量建立了诊断模型。在现有神经网络故障诊断方法基础上,提出了一种基于带有偏差单元递归神经网络的在线故障诊断方法,设计了相应的故障样本和故障编码。仿真结果表明,该方法在收敛速度、非线性能力及精度方面明显优于一般方法。对故障模式的回想结果及实际运行结果证明,本方法切实可行,适合于旋转机械的在线故障诊断。

Fault Diagnosis for Rotating Machinery Based on RNN

In order to diagnose faults of rotating machinery on-line,spectrum analysis is used to analyze the information from fuselage vibrations of ten kinds of faults,the result indicates that there exist various fault information.The fault diagnosis model is built by using the fault eigenvectors.Based on the existing neural network fault diagnosis methods,a new on-line fault diagnosis method using the recurrent neural network with deviation error units is presented,the fault samples and coding are designed.Simulation shows the method is better than usual methods in converging speed,non-linear and precision.Testing and recollection results of fault vectors show the method is effective.