改进的LVQ神经网络在风机故障诊断中的应用

提出一种改进的LVQ神经网络的风机故障诊断新方法。利用风机振动频域的特征向量作为学习样本,建立与风机故障类型的映射关系。将能量特征输入改进的LVQ神经网络进行网络训练与检测,以实现风机的故障识别。经比较,其性能优于BP网络和遗传网络,诊断正确率高达96%以上。通过仿真实验和风机的故障诊断实例表明:该网络提高了收敛速度及诊断精度,有效地抑制网络陷于局部极小,更适合风机等较复杂分类问题的故障诊断。

Application of Improved LVQ Neural Network in Fault Diagnosis of Fans

The fan＇s new fault diagnosis which employs the improved learning vector quantization（LVQ） neural network was proposed,in which,the feature vector of the fan vibration in frequency domain was taken as learning samples to reflect mapping relationship of fault types,then the energy feature was inputted to train LVQ neural network so that the fan＇s faults can be recognized.The comparison results show that the over 96%accuracy of the improved LVQ network outperforms that of both BP network and genetic network.Both simulation results and examples of the fan fault diagnosis show that this improved LVQ neural network can enhance convergence speed and diagnosis accuracy and effectively repress the phenomenon of trapping in the local minimum for the network,as well as satisfy the fan fault diagnosis and other complex classification problems.