人工神经网络在故障诊断系统中的应用

人工神经网络以其独到的联想、记忆、储存和学习功能，在故障诊断领域受到了广泛关注。其中，BP网络是最成熟、应用最广泛的一种网络。但BP网络在实际应用中也存在诸多不足，如学习率的取值需要凭借经验或试算选取，网络学习收敛速度慢以及易陷入局部最小点等。针对BP网络的这些不足，采用了Rumelhan附加惯性冲量和动态调整学习率相结合的改进的BP网络方法，并尝试用该改进算法来对凝汽器进行故障诊断。事实证明，在平均情况下，本系统收敛速度远远快于基本的BP算法。

Application of Artificial Neural Networks in Fault Diagnose System

Artificial Neural Networks(ANN)has broadly attracted the attention of the scientists that are in fields of fault diagnoses, for its particular functions of association, memory, storage and learning. Back Propagation(BP)networks are the most mature and the most broadly used network of ANN. But BP networks still have numerous flaws in practice. Such as learning rates have to be valued by experiences or experimental calculations; its' rate of learning convergence is too slow; it frequently falls into local minimum points while learning, etc. In this paper, aiming at dealing the flaws above, it synthesizes methods of Rumelhart's add-in momentum item and dynamically adjusting learning rates to ameliorate BP networks. And it attempts to use the improved BP methods to diagnose vapor congealing device .On the average situation, it's been proved that the convergent rate of this system is much faster than the basic BP algorithm.