A Bayesian network-based approach for the assessment and management of ageing in major hazard establishments

Currently, there is an increasing attention towards ageing of industrial equipment, as the phenomenon has been recognised as a cause of severe accidents, recorded in the last years in many process establishments. Recent studies described ageing through a number of key-factors affecting the phenomenon by accelerating or slowing it down. The Italian Competent Authority for the prevention of chemical accidents (Seveso III Directive) adopted a short-cut method, accounting for the assessment of these factors, to evaluate the adequateness of ageing management during inspections at Seveso sites. In this paper, a Bayesian Network was developed, by using the data gathered during the first application of the short-cut method, with the aim to verify the robustness of the approach for ageing assessment and the validity of the a priori assumptions used in assessing the key-factors. The structure of the Bayesian network was established by using experts’ knowledge, whereas the Counting Learning algorithm was adopted to execute the parameter learning by means of the software Netica. The results showed that this network could effectively explore the complex logical and uncertain relationships amongst factors affecting equipment ageing. Results of the present study were exploited to improve the short-cut method.     

How can process safety and a risk management approach guide pandemic risk management?

The coronavirus disease (COVID-19) brought the world to a halt in March 2020. Various prediction and risk management approaches are being explored worldwide for decision making. This work adopts an advanced mechanistic model and utilizes tools for process safety to propose a framework for risk management for the current pandemic. A parameter tweaking and an artificial neural network-based parameter learning model have been developed for effective forecasting of the dynamic risk. Monte Carlo simulation was used to capture the randomness of the model parameters. A comparative analysis of the proposed methodologies has been carried out by using the susceptible, exposed, infected, quarantined, recovered, deceased (SEIQRD) model. A SEIQRD model was developed for four distinct locations: Italy, Germany, Ontario, and British Columbia. The learning-based approach resulted in better outcomes among the models tested in the present study. The layer of protection analysis is a useful framework to analyze the effect of different safety measures. This framework is used in this work to study the effect of non-pharmaceutical interventions on pandemic risk. The risk profiles suggest that a stage-wise releasing scenario is the most suitable approach with negligible resurgence. The case study provides valuable insights to practitioners in both the health sector and the process industries to implement advanced strategies for risk assessment and management. Both sectors can benefit from each other by using the mathematical models and the management tools used in each, and, more importantly, the lessons learned from crises.     

Application of dynamic risk analysis in offshore drilling processes

Process safety is the common global language used to communicate the strategies of hazard identification, risk assessment and safety management. Process safety is identified as an integral part of process development and focuses on preventing and mitigating major process accidents such as fires, explosions, and toxic releases in process industries. Accident probability estimation is the most vital step to all quantitative risk assessment methods. Drilling process for oil is a hazardous operation and hence safety is one of the major concerns and is often measured in terms of risk. Dynamic risk assessment method is meant to reassess risk in terms of updating initial failure probabilities of events and safety barriers, as new information are made available during a specific operation. In this study, a Bayesian network model is developed to represent a well kick scenario. The concept of dynamic environment is incorporated by feeding the real-time failure probability values (observed at different time intervals) of safety barriers to the Bayesian network in order to obtain the corresponding time-dependent variations in kick consequences. This study reveals the importance of real-time monitoring of safety barrier performances and quantitatively shows the effect of deterioration of barrier performance on kick consequence probabilities. The Macondo blowout incident is used to demonstrate how early warnings in barrier probability variations could have been observed and adequately managed to prevent escalation to severe consequences.     

基于节点流量校核的供水管网爆管定位研究

针对目前基于节点压力变化定位供水管网爆管的方法,当供水管网中泵切换时,导致定位爆管位置存在误报的问题,提出了基于节点流量校核的管网爆管定位方法.利用奇异矩阵法推导出节点流量灵敏度矩阵,采取奇异矩阵中的最优搜索向量计算方法找到节点流量变化位置,确定爆管管道.首先,将该方法应用于一个小管网,阐述基于节点流量校核的管网爆管定...     

A hybrid model for human factor analysis in process accidents: FBN-HFACS

Human factors are the largest contributing factors to unsafe operation of the chemical process systems. Conventional methods of human factor assessment are often static, unable to deal with data and model uncertainty, and to consider independencies among failure modes. To overcome the above limitations, this paper presents a hybrid dynamic human factor model considering Human Factor Analysis and Classification System (HFACS), intuitionistic fuzzy set theory, and Bayesian network. The model is tested on accident scenarios which have occurred in a hot tapping operation of a natural gas pipeline. The results demonstrate that poor occupational safety training, failure to implement risk management principles, and ignoring reporting unsafe conditions were the factors that contributed most failures causing accident. The potential risk-based safety measures for preventing similar accidents are discussed. The application of the model confirms its robustness in estimating impact rate (degree) of human factor induced failures, consideration of the conditional dependency, and a dynamic and flexible modelling structure.     

基于随机森林的阻变型通风网络故障诊断方法*

为了在矿井通风网络发生阻变型故障时,能够快速准确判断出故障位置和故障量,提出1种基于随机森林的通风网络故障位置和故障量诊断方法。利用矿井通风仿真系统IMVS将唐安矿模拟故障生成空间数据集并进行数据预处理,构建基于随机森林的故障诊断模型,并利用该诊断模型对唐安矿矿井通风网络模拟故障位置和故障量进行判断和预测。引用多种方法对模型进行度量,通过唐安矿模拟实验验证基于随机森林的故障诊断模型的有效性。将随机森林和决策树的故障诊断准确率进行对比,研究结果表明:随机森林较决策树故障准确率有进一步的提高,并发现故障地点失误诊断多是相邻巷道,在一定程度上工作人员对故障地点的判断并不受其影响。     

基于Bayesian网络的常减压装置设备故障概率分析

为分析影响常减压蒸馏装置平稳运行的设备失效模式及故障部件,基于1 151条设备故障数据,采用Bayesian网络分析方法,分别对离心泵、压缩机、电动机构建基于Bayesian网络的设备故障概率分析模型,分析故障部件、失效模式、故障后果之间的定量概率关系。研究结果表明：离心泵、压缩机、电动机停运的关键致因部件分别为轴承箱密封故障、活塞环故障、轴承故障,同时得到导致设备停运的故障部件敏感度排序。研究结果有助于提高设备故障风险防范及检维修工作效率,同时可为备件优化方案提供思路。     

基于HFACS-BN模型的危险化学品事故人因路径分析*

为提高危险化学品安全管理水平,收集我国134起以人为主因的危险化学品事故,构建危险化学品事故的HFACS-BN模型,并基于贝叶斯网络对危险化学品事故中的人因路径及各因素灵敏度进行分析。研究结果表明:行为违规是导致事故的最主要不安全行为;在复杂路径中,环境因素和操作者状态具有较高的灵敏度,与其相关的风险认知与处理不当是阻碍事故预防的关键因素;在非复杂路径中,操作者状态与运行计划不当具有较高的灵敏度,其所涉及的监督违规、组织过程和组织氛围,是导致危险化学品事故的根本起点,并可依此分析事故直接原因和间接原因;操作者状态在2种路径中均表现出高灵敏度,因此化工企业在人员管理上要加强关注职工状态,减少不安全行为的出现。     

城市燃气管网泄漏事故分析知识图谱构建及应用研究*

为深入认识燃气管网泄漏事故的发生发展机理,提高事故分析预测的自动化、智能化、数字化水平,利用知识图谱对燃气管网泄漏事故进行研究。在事故案例分析的基础上,从人-物-环-管的角度对燃气泄漏过程以及火灾爆炸次生事故的相关实体进行归纳梳理,对实体间的逻辑关系和非逻辑关系进行辨识,并对实体的属性进行分类,进而构建出较为全面的燃气管网泄漏事故知识图谱。在此基础上,搭建BP神经网络模型,基于已知实体或属性状态,预测相关联其他实体或属性的状态。研究结果表明:燃气管网知识图谱能够有效展示燃气管网泄漏事故发展的动态过程及相关要素,结合BP神经网络能够有效预测事故的发展路径及相关状态,从而提高燃气管网泄漏事故的分析预测水平与效率。     

基于卷积神经网络的冲击地压预警方法研究*

为探究深度学习在冲击地压预警方面的应用前景,以新疆某冲击地压矿井为研究背景,将深度学习和专家评判运用到微震数据分析中,基于卷积神经网络构建冲击地压预警模型。充分利用一维卷积神经网络对时序数据有较强特征提取能力的优势,以微震数据及其特征参数作为输入,以专家评判值作为标签,借助Python-Keras框架实现冲击地压预警模型的构建和训练。研究结果表明:模型预警效果并不随着训练迭代次数的增加而逐渐最优,存在最优迭代次数,对于所建模型当迭代次数为30时测试集的冲击危险预测结果与专家评判结果基本吻合,同时说明模型可以较好地学习专家评判经验实现冲击地压预警。研究表明所建模型对研究时段内发生的5次大能量矿震事件均进行预警,其准确度较高,具有现场实际应用价值。