ELM在航空发动机气路部件故障诊断的应用研究

Research on ELM Application in Aeroengine gas Path Components Fault Diagnosis

针对航空发动机故障呈现复杂性、多样性、非线性等特点,运用传统的BP、ELman神经网络进行发动机气路部件故障诊断存在网络训练参数设置复杂,迭代次数多,训练速度慢,泛化能力欠缺等问题。为此提出利用极限学习机(Extreme Learning Machine,ELM)法识别涡扇发动机气路部件故障。该方法通过为输入权和隐藏层偏置随机赋值,利用MP逆求解输出权值。以某型涡扇发动机为对象,进行ELM、BP、ELman气路部件故障诊断比较研究,实验结果验证了利用ELM识别涡扇发动机气路部件故障的精确性、快速性、稳定性。

The aeroengine gas path component faults have the characteristics of complexity, diversity and nonlinear. When applied traditional BP and ELman into aeroengine gas path components fault diagnosis, there exist some problems as complex setting of network parameters, multiple iterations, slow training speed and poor generation performance. In order to solve these problems, Extreme Learning Machine （Extreme Learning Machine, ELM）method was used to aeroengine gas path components fault diagnosis This method randomly assigned input weights and biases and calculate output weights by Moore-penrose, pseudoinverse. ELM,BP and ELman were studied and compared on a certain type of turbofan engine The results verify the diagnostic accuracy, quickness and stability of ELM.