Abstract:

As a knowledge representation model, fuzzy Petri nets can be potentially used in industrial process systems for fault reasoning and diagnosis. The establishment of fuzzy Petri nets model usually demands for a priori knowledge, which discourages the use in practice. To utilize industrial process data effectively, association rules based on conditional state fuzzy Petri nets are proposed, which are subsequently applied to industrial process fault reasoning and diagnosis. Fuzzy rules along with confidences of fuzzy Petri nets are extracted by association rule algorithms of data mining. Key principal components (conditional variables) affecting the confidence are extracted by correlation analysis between variables and conditional states, thus creating the conditional state fuzzy Petri nets. The reverse reasoning of dynamic confidence is performed with the iterative algorithm of the maximal algebra, obtaining the probability of fault occurrence in industrial processes. This approach realizes data driven fault diagnosis, so as enhancing the speed and accuracy of fault diagnosis. A case study on chemical reactions shows the effectiveness of the proposed method.

Key words: process systems, association rules, conditional state fuzzy Petri nets, dynamic confidence, reverse reasoning, fault diagnosis, chemical reaction

摘要：

模糊Petri网作为一种知识表达模型，能够用于工业过程系统故障推理和诊断。然而，模糊Petri网的建立大多需要先验知识，为此限制了其广泛应用。为了能够有效利用工业生产过程数据，提出了一种基于关联规则的条件状态模糊Petri网，并将其用于工业过程故障推理与诊断。采用数据挖掘的关联规则方法提取模糊Petri网的模糊规则及置信度，通过变量间的关联分析，将影响置信度的关键主元（条件量）提取出来，建立条件状态模糊Petri网；基于极大代数的迭代算法进行动态置信度逆向推理，可以获得工业过程的故障发生概率。该方法实现了故障诊断网络的数据驱动，从而提高故障诊断的快速性与准确性，某化学反应研究表明所提方法的有效性。

关键词: 过程系统, 关联规则, 条件状态模糊Petri网, 动态置信度, 逆向推理, 故障诊断, 化学反应