Abstract: At present, the essential reliability of chemical equipment has been increasingly improved with a development of automation and information technology, and human error has become the main factor over chemical accidents. Hence it is necessary to conduct human reliability analysis (HRA) on chemical systems to ensure their safe and stable operations. However, the chemical systems have a high non-linearity and complex coupling. The existing HRA technologies cannot be directly applied on them, and the effectiveness and accuracy of analysis results are poor. Therefore, this work proposes a structured assessment method of human error probability for chemical systems. This method particularly combines the human-hazard and operability analysis (human-HAZOP) with Bayesian network (BN). Firstly, according to the operation manual, process flow diagram, and other information of a chemical system, the tasks are summarized for human-HAZOP and further subdivided into several operation steps and behaviors. Secondly, meaningful deviations are selected using the operation behaviors and guide words. Accordingly, the potential causes, possible consequences, existing measures, and suggested measures are analyzed to form a structured human-HAZOP report for this chemical system. Thirdly, based on the human-HAZOP results, the potential causes and possible consequences are considered as leaf nodes and root nodes, respectively. A structured BN model is finally established to calculate the human error probability in the chemical system. The esterification reaction of one methyl-acrylate virtual simulation factory is taken as a test case. The probability of human error for the esterification reaction is calculated to be 0.0004 and the main human error behavior is maintenance staff stealing work omission of maintenance staff. Traditional CREAM analysis of this chemical system results in human error probability results of 0.0001 to 0.01. Results show that this structured method can effectively and accurately assess the human error probability of chemical systems in comparison with the traditional CREAM.

Key words: Chemical systems, Human error probability, Structured assessment, Human-hazard and operability analysis (Human-HAZOP), Bayesian network

摘要： 化工生产系统高度非线性且复杂耦合化，使现有人因可靠性分析(HRA)技术难以直接应用，且分析结果的有效性和准确性差。为此，本工作将人因-危险和可操作性分析(人因HAZOP)与贝叶斯网络(BN)相融合，提出了一种结构化的化工系统人因失误概率评估方法。通过概述人因HAZOP的作业任务，筛选引导词和偏差，形成结构化的人因HAZOP报告，并据此构建BN模型，求解化工系统人因失误概率。以丙烯酸甲酯虚拟仿真工厂的酯化反应为例，人因失误概率的计算结果为0.0004，且主要人误行为是维护人员遗漏。对该化工系统进行传统CREAM分析，人因失误概率结果为0.0001~0.01。与传统CREAM相比，这种结构化方法能够高效且精确地评估化工系统人因失误概率。

关键词: 化工系统, 人因失误概率, 结构化评估, 人因-危险和可操作性分析（人因HAZOP）, 贝叶斯网络