基于自联想网络的发动机传感器解析余度技术

ANALYTICAL REDUNDANCY BASED ON AUTOASSOCIATIVE NEURAL NETWORK FOR AEROENGINE SENSORS

本文提出了一种基于自联想神经网络的传感器解析余度技术。在这种网络中，冗余传感器的信息被压缩、重组进入网络的第一部分，网络的第二部分将压缩信息恢复出来。基于数据融合原理，若一个传感器发生故障，其它传感器仍可提供足够的信息代替发生故障的传感器。理论分析和用于涡轴发动机的仿真结果表明，这种特殊结构的自联想网络具有良好的过滤噪声和故障信号的作用。

Analytical redundancy technology based on autoassociative neural network is presented for aeroengine sensors.It doesn't depend on model and needs only the sample net of measurement data without sensor fault to train the network.Then it can work on-line.Estimation Feedback Scheme is developed and can meet the real-time requirements.If there has been performance degeneration of aeroengines,a compensation algorithm can be used automatically.Sensor fault analytical redundancy is accomplished by integrating the network estimation and fault detection logic.The results of theoretical analysis and simulation show that the provided scheme has the ability of distinguishing performance degeneration and sensor faults,and it also can detect soft faults of sensors.