加快神经网络训练速度的方法在旋转机械故障诊断中的应用

FAULT DIAGNOSIS OF ROTATING MACHINERY USING BACK PROPAGATION NEURAL NETWORK

根据一双跨转子实验台，模拟了转子与静子在轮盘处及轴颈处碰磨、轴系不对中及转子不平衡故障，通过一个信号自动处理装置记录下转子正常振动信号及发生各种故障时的信号，然后利用研制的人工神经网络系统对故障示例进行学习。通过在实际中诊断故障，证明这种是可行的。本文还针对人工神经网络（ＢＰ算法）存在的训练速度慢的问题，提出了一个加快网络训练速度的新方法（ＡＲＢＰ算法），较大提高了网络的训练速度。

The faults of rub between rotor and stator,misalignment test rig of the shafting,and unbalance of the rotor are simulated at a compressor test rig.The normal vibration signals of the rotor and the signals of the various emerging faults are registered and sampled by an automatic signal proceessor.Then a new-developed back propagation neural network is employed to learn the fault diagnosis from the sampled faults and a new method for accelerating the learning speed has been proposed in this paper.