基于NN-ELM的航空发动机燃油系统执行机构故障诊断

提出了一种航空发动机执行机构及其传感器单一故障诊断及定位方法.首先通过执行机构模型判断是否发生故障,然后运用发动机逆模型对故障进行定位.基于离线训练BP(back propagation)神经网络建立执行机构模型,根据某半物理仿真试验台的测试数据训练网络参数.提出离线训练和在线训练相结合的极端学习机(ELM)算法建立发动机逆模型,使网络在初始时刻就具有诊断能力,工作过程中具有适应能力,且在线训练过程采用阈值判别法筛选训练样本,减小了在线训练时间,提高了逆模型的实时性.以某型发动机燃油系统执行机构为例的设计和仿真结果表明:该诊断系统能够准确地对发动机在稳态和动态工况以及蜕化状态下的执行机构及其传感器单一故障进行准确诊断和定位,具有很好的实时性. 

Fault diagnosis for actuators of aero-engine fuel system based on NN-ELM

A single fault diagnosis and location method of aero-engine actuator and its sensor was proposed. Fault of actuator was diagnosed according to the actuator model and located according to the inverse engine model. The actuator model was built based on offline training BP(back propagation) neural network (NN). The network parameters were trained according to the test data of a semi-physical simulation test bed. The inverse engine model based on offline and online extreme learning machine (ELM) had original diagnosis ability and adaptability during operation. The training samples were selected using the threshold discrimination, which can reduce the online training time and improve the real-time property of inverse model during the online training process. Design and simulation results of a certain engine fuel system actuator show that,the proposed system can diagnose and locate the single faults of engine actuator or its sensor accurately in steady state, dynamic state and degradation state. The real-time property can be satisfied.