一种基于人工神经网络的旋转机械故障监测方法

A Rotary Machine Fault Monitoring Method Based Artificial Neural Network

把神经网络与数字信号处理相结合，提出并实现了一种基于神经网络的旋转机械在线故障监测系统，主要用于识别旋转机械的一些常见故障如转子不平衡振动、转子零部件松动、油膜涡动等等。系统主要由数字信号采集、数字信号预处理和Ｂ－Ｐ神经网络几部分组成。在神经网络中采用批处理加动量项的梯度下降算法学习。仿真结果表明该故障监测系统在故障识别率等各方面性能指标均优于传统的基于知识的工况监测与故障诊断系统。

Integrating neural network with digital signal process, an rotary machine on-line fault monitoring system using neural network is proposed and realized in this paper, and this system is mainly used to identify some of rotary machine’s faults such as rotor unbalanced vibration, rotor looseness, the whirlpool of the oil’s film, etc. This system consists of a sensor part, a gathering apparatus part, a preparing part of process the digital signals and the back-propagation network part. A weight-learning algorithm is presented using the gradient descent method with batch process and adding a momentum item. Simulation results show that the proposed fault diagnostic system has better performance than the traditional state monitoring and fault diagnostic system based on knowledge in many aspects such as the discriminating rate.