基于动态SSL的航空发动机突发故障检测

Abrupt Failure Detection for Rotor System of Aero-Engine Based on Dynamic Semi-Supervised Learning

针对航空发动机转子系统的突发故障检测,提出了一种适用于动态过程检测的半监督学习(semi-supervisedlearning,简称SSL)方法———动态半监督学习方法(dynamic semi-supervised learning,简称DSSL)。首先,用已经有类标的数据对学习器模糊KNN(fuzzy k-nearest neighbour,简称FKNN)进行初始化训练,训练完成后,当新的数据到达时,对新的数据进行分类;然后,计算类的演化指标来检测系统的演化程度,检测阶段完成后,学习器根据检测结果实时的修正自身参数,以自适应最终的动态分类任务;最后,用转子试验台模拟航空发动机突发性扇叶断裂故障来获取数据,该实验结果验证了提出方法在突发性故障检测中的可行性、有效性。

To detect problems for the abrupt failure of the aircraft engine rotor system,dynamic semi-supervised Learning(DSSL) algorithm is proposed based on the fuzzy k-nearest neighbour method for the dynamic evolution of the system detects.In the first phase,the labeled training data is used to initialize the classifier for FKNN learning.Then in the second phase a class evolution can be detected and be confirmed after the classification of each new pattern.In the last phase,the parameters of the evolved class are updated incrementally.Finally,the approach is illustrated feasibility and validity using the data which obtain from the rotor test stand simulation of aero-engine abrupt blade fracture failure.