基于神经网络的信息融合故障诊断技术

利用神经网络技术建立信息融合中心, 对多传感器数据进行融合处理, 通过多源互补信息减小故障诊断系统的不确定性。文中讨论了神经网络多传感器信息融合方法中数据预处理与特征提取、特征向量维数压缩与关联、归一化处理方法等, 同时, 对神经网络的构造以及学习训练等内容, 也作了较为详细的讨论。通过对柴油机振动监测数据、燃油压力波动信息、以及两者融合信息的故障诊断性能的比较, 表明神经网络多传感器信息融合方法用于复杂机械的故障诊断是可行和有效的。
关　键　词: 多传感器; 信息融合; 神经网络; 故障诊断
Abstract: Th is paper deals w ith a mult isenso r system app lied to mechanical fault diagno sis. By using the info rmat ion fusion pat tern based on neural netwo rk, the uncertainty of the diagno sis system can be decreased great ly. The methods fo r data p rep rocessing, features ext ract ion and const ruct ion of the neural netwo rk are discussed. The app roach is show n to be p ract ical and effect ive th rough diagno sing the faults in the fuel inject ion system of a diesel engine.
Key words:M ult isenso r; Info rmat ion fusion;N eural netwo rk; Fault diagno sis