基于概率神经网络的离心式制冷机故障诊断
Fault diagnosis for centrifugal refrigeration system based on probabilistic neural network
摘要：
使用概率神经网络（PNN）对制冷系统7种常见故障进行诊断，包括系统故障和局部故障。详细介绍了应用PNN建立故障诊断模型以及平滑因子寻优过程，并探索了样本规模对最佳平滑因子和诊断正确率的影响。将PNN与人工神经网络中最常用的误差反向传播（BP）神经网络进行比较，结果表明，PNN网络的诊断正确率比BP网络诊断正确率高3.48%，且诊断耗时更短，并且PNN网络的单次训练结果更可靠。尽管2种网络的训练结果均显示系统故障比局部故障更难以被识别，但使用PNN网络进行诊断时，系统故障的诊断正确率明显高于BP网络的诊断正确率。
Abstract：
Applies the probabilistic neural network (PNN) to diagnose seven types of typical faults for a refrigeration system, including system-level faults and component-level faults. Elaborates the establishment of the fault diagnosis model based on PNN and the optimal processes of finding out the best spread value in detail. Studies the influence of sample size on the best spread value and the correct diagnose rate. Compares the performance of the PNN and the prevailing back-propagation (BP) neural network. The results show that the overall correct diagnosis rate of the PNN model is 3.48% higher than that of the BP network, which consumes much less diagnosing time, and the diagnosis of single training with the PNN is more reliable than that of the BP network. Although the diagnosis results of these two networks show that the system-level faults is more difficult to be identified than the component-level faults, great improvement still has been observed by using PNN.
Keywords:faultdiagnosis,refrigerationsystem,spreadvalue,probabilisticneuralnetwork(PNN),back-propagation(BP)network