航空发动机油液磨粒与故障关系RVM预测方法

Research and Application on A Predictable Method of the Relationships between Aero Engine Oil Wear Particle and Fault

支持向量机方法应用于发动机油液磨粒与故障关系预测时,常出现稀疏性不强、计算量大、核函数必须满足Mercer条件等问题。针对这一问题,在原理介绍和推导的基础上,利用某型航空发动机滑油光谱分析得到的237组数据中的7种金属磨粒浓度及其对应的发动机工作状态,对相关向量机预测方法进行了检验。采用相关向量机、最小二乘支持向量机和反向传播神经网络方法,对发动机工作状态进行预测。结果表明,在同等条件下,与LSSVM和BP-NN相比,RVM拥有计算量较少、预测时间较短,精度较高等优势,可广泛应用于发动机油液磨粒分析与故障预测。

In order to overcome some inherent defects of Support Vector Machine （SVM）,such as poor sparsity,heavy computation and kernel function satisfactory to the Mercer＆#39;s conditions in the engine oil wear particle detection and the fault diagnosis,a new attempt by adopting a predictable method based on Relevance Vector Machine （RVM）is proposed.On the basis of the introduction of principle and deduction, the spectrum analysis data of a certain aero engine lubricating oil are utilized to predict the relationships between the aero engine oil wear particle concentration and fault.Through analysis and verification,the results show that the method based on RVM has more advantages in generalization over the SVMs and the ANNs under the same conditions,and the method can be widely used in the engine oil wear particle analysis and the failure prediction.