贝叶斯网络在生产过程故障诊断中的应用

摘要

针对单一故障诊断方法的不足，提出了一种综合不同诊断方法的基于贝叶斯网络( BN)的故障诊断方案。该方案在构建贝叶斯网络结构时，除了将反映因果关系的过程变量作为节点外，同时将单一故障诊断方法对过程故障的诊断率作为证据节点添加到网络中。采用一种基于均值属性分析的贝叶斯统计学的学习算法来确定已知结构的后验概率参数，从而实现参数学习。最后通过贝叶斯推理得出故障原因的概率。以Tennessee Eastman过程的故障诊断为例，通过仿真试验证明了基于BN的故障诊断方案能综合不同故障诊断方法的优点，有效提高了诊断率。%Against the deficiency of single fault diagnosis method, the fault diagnosis method based on Bayesian network ( BN) and integrating different diagnosis methods is proposed. When establishing BN, besides the process variables which reflecting the causality, the diagnosis rate of the single fault diagnosis method is also added in the network as the evidence node. By using Bayesian statistical learning algorithm based on mean attribute analysis, the posteriori probability parameters of known structure are determined, thus the parameter learning is implemented. Finally, the probability of causes of the failure is obtained through Bayesian inference. With fault diagnosis of Tennessee Eastman (TE) process as example, through emulated experiment, it is proved that the strategy proposed integrates the advantages of different fault diagnosis methods, and effectively enhances the diagnosis rate.