航空发动机传感器解析余度模型的建立方法

为了使用解析余度模型对传感器故障进行诊断,提出了1种基于K-均值聚类与改进微分进化算法优化的极端学习机(IDE-ELM)的发动机传感器解析余度模型建立方法。为避免求解ELM算法时H矩阵奇异,采用K-均值聚类对试验数据进行聚类处理,然后从每类数据中选取1组数据组成训练样本用于训练;利用IDE算法优化ELM的输入层权值和偏置,提高ELM的泛化能力。利用飞行试验数据进行了仿真验证。结果表明:基于K-均值聚类和IDE-ELM设计的传感器解析余度模型具有较高的精度,可用于FADEC系统双通道传感器的故障诊断。

Modeling Method of Analytical Redundancy Model of Sensors for Aero-engine

In order to diagnose sensor fault using analytical redundancy model, a modeling method for analytical redundancy model of sensors was proposed based on K-means clustering and extreme learning machine (ELM) optimized by improved differential evolution (IDE) algorithm. To avoid the H matrix singularity during solving the ELM algorithm, K-means clustering was used to cluster the test data, and then a set of data was selected from each kind of data to form training samples. The IDE algorithm was used to optimize the input layer weight and bias of ELM, which could improve the generalization ability of ELM algorithm. The simulation experiments using flight test data was carried out. The results show that the established analytical redundancy model of sensor based on K-means and IDE-ELM achieves high accuracy and can be used to dual-channel sensors diagnosis