Abstract:

Fault diagnosis techniques based on probabilistic graphical models are often used for uncertain information reasoning. Among them, Bayesian network, an effective tool which has strong characteristics of practicality and applicability, is used extensively. For tracking fault propagation paths in industrial systems with large volume data, this paper proposes a Bayesian-network-based algorithm for fault path tracking. Given all possible combinations of parents, values of specific symptom child node with the maximum conditional probability are estimated by using   decomposition of conditional probability as well as Dichotomy after Bayesian network training. Then the most probable cause is determined by comparing estimations with the observed data. Through layer-by-layer analysis, fault propagation paths in the network can be finally tracked. The experimental results demonstrate that the novel approach is capable of tracking fault paths effectively, which provides higher adaptability and faster speed.

Key words: Bayesian network, fault path tracking, probability reasoning, maximum conditional probability

摘要：

基于概率图模型的故障诊断技术常被用于不确定信息的推理。其中贝叶斯网络作为一类具有强大实用性及适用性的概率图模型，得到了广泛应用。为有效识别大数据量工业系统中的故障传播路径，提出了一种考虑父节点影响的贝叶斯网络故障路径追溯算法。在建立贝叶斯网络后，给定父节点所有潜在取值组合，算法利用条件概率分解及二分法估计特定故障子节点的最大条件分布值，然后与真实观测值比较，得到导致此故障发生的最大可能原因。通过逐层推断，最终能够实现对网络中故障传播路径的有效追溯。实验结果表明，该算法具有准确追踪故障路径的能力，同时显示出较高的可用性及较快的追踪速度。

关键词: 贝叶斯网络, 故障路径追溯, 概率推理, 最大条件概率