摘要
对离心式空气压缩机动不平衡的故障问题 ,采用神经网络的BP (BackPropagation)算法进行故障模式识别和诊断 ,并针对传统BP算法收敛速度慢 ,宜陷入局部最小的情况 ,从以下方面进行处理 :其一 ,使用带动量改进型反向传播BP算法加快了收敛速度 ;其二 ,训练过程中对隐层和输出层采用了双曲正切阈值激励函数进行训练 ,解决了Sigmoid函数在 0和 1附近易陷入平坦区的情况。成功实现了故障样本空间到诊断数据空间的影射 ,并且在理论上给出了数学推导。
In this article,a new improving method for BP neural network is used to fault identification and diagnosis of compressor.As to BP algorithmic's shortcomings of slow convergence rate and proneness to yield minimal local results,the improving method has the following characteristics:Firstly,convergence rate of BP algorithm with momentum is faster than the standard BP algorithm.Secondly,in order to overcome sigmoid function incidental to lead to proneness to yield minimal local results in the train course of the hidden unit and output unit,which is tanh(x) a new threshold activation function is applied.Lastly,the mathematical speculated of BP algorithm is given.
出处
《压缩机技术》
2002年第5期1-4,共4页
Compressor Technology
关键词
BP算法
神经网络
动不平衡
离心式空气压缩机
故障诊断
BP algorithm
neural network
dynamic imbalance
centrifugal air compressor
fault diagnosis
