Learning from accidents – What more do we need to know?

A model of experience feedback (the CHAIN model) that emphasizes the whole chain from initial reporting to preventive measures is used to identify important research needs in the field of learning from accidents. Based on the model, six quality criteria for experience feedback after an accident or incident are presented. Research on experience feedback from accidents is reviewed. The overall conclusion is that the discipline of experience feedback has not been sufficiently self-reflective. The process of experience feedback can and should be applied to experience feedback itself, but that is rarely done. Evaluation studies are needed that provide hard (evidence-based) information about the effects of various methodologies and organizational structures. Four types of studies are particularly important for the development of evidence-based accident investigation practices: (1) studies of the effects and the efficiency of different accident investigation methods, (2) studies of the dissemination of conclusions from accident investigation, (3) follow-up studies of the extent to which accident investigation reports give rise to actual preventive measures, and (4) studies of the integration of experience feedback systems into overall systems of risk management.     

PDE accident model from a safety information perspective and its application to Zhangjiakou fire and explosion accident

Accident models can provide theoretical frameworks for determining the causes and mechanisms of accidents, and thus are theoretical bases for accident analysis and prevention. The role of safety information in accident causation is profound. Thus, safety information is an important and essential perspective for developing accident models. This study presents a new accident model developed from a safety information perspective, called the Prediction—Decision—Execution (PDE) accident model. Because the PDE accident model is an emerging accident model that was proposed in 2018, its analysis logic and viability remain to be discussed. Thus, the main contributions of this study include two aspects: (i) detailed explanation of the analysis logic of the PDE accident model, and (ii) case-study examination of the Zhangjiakou fire and explosion accident, a serious accident that occurred in China in 2018, to demonstrate the viability of the PDE accident model. Results show that this is a safety-information-driven accident model that can provide a new and effective methodology for accident analysis and prevention, and safety management.     

The development of a more risk-sensitive and flexible airport safety area strategy: Part I. The development of an improved accident frequency model

This two-part paper presents the development of an improved airport risk assessment methodology aimed at assessing risks related to aircraft accidents at and in the vicinity of airports and managing airport safety areas (ASAs) as a risk mitigation measure. The improved methodology is more quantitative, risk-sensitive, flexible and transparent than standard risk assessment approaches. As such, it contributes to the implementation of Safety Management Systems at airports, as stipulated by the International Civil Aviation Organisation.The first part of the paper presents the methodological advances made in the development of accident frequency models; namely the building of a single comprehensive database of all relevant accident types, the collection and use of normal operations data in quantifying the criticality of a series of risk factors, and modelling accident frequency using multivariate logistic regression. The resulting models have better goodness-of-fit, sensitivity and specificity than standard risk assessment methodologies.     

Accident modelling and analysis in process industries

Accident modelling is a methodology used to relate the causes and effects of events that lead to accidents. This modelling effectively seeks to answer two main questions: (i) Why does an accident occur, and (ii) How does it occur. This paper presents a review of accident models that have been developed for the chemical process industry with in-depth analyses of a class of models known as dynamic sequential accident models (DSAMs). DSAMs are sequential models with a systematic procedure to utilise precursor data to estimate the posterior risk profile quantitatively. DSAM also offers updates on the failure probabilities of accident barriers and the prediction of future end states. Following a close scrutiny of these methodologies, several limitations are noted and discussed, and based on these insights, future work is suggested to enhance and improve this category of models further.     

Towards an evaluation of accident investigation methods in terms of their alignment with accident causation models

The purpose of this paper is to reflect on accident causation models and accident investigation methods. Theories on accident causation and the modelling of accident mechanisms, as well as a number of methods for accident investigation have been developed and described in the literature. The evolution of accident causation models over time shows a shift from the sequence of events to the representation of the whole system. Respectively, the evolution of accident investigation methods over time reveals a gradual shift from searching for a single immediate cause, to the recognition of multiple causes. In order to evaluate the accident investigation methods, specific plausible requirements were established in order to verify that a specific accident investigation method fulfils the principles of a specific accident causation model or give evidence to the degree of alignment between them. Since different models approach accident causation in different ways, methods linked to these models provide fragmentary information regarding the accident. It is therefore expected that using a combination of model-method pairs could provide a more reliable platform for accident analysis.     

A modified accident analysis and investigation model for the general aviation industry: Emphasizing on human and organizational factors

IntroductionCurrently, there is a lack of specific analytical tools for general aviation accidents (GAAs). This has led to loopholes in the prevention of GAAs.MethodsA Swiss Cheese model for general aviation (SCM-GA) is proposed to identify the human and organizational factors involved in GAAs. In the proposed SCM-GA, 5 categories, 45 subcategories, a general aviation safety management system (GA-SMS) and safety culture were developed based on the classic accident causation models combined with the laws and regulations and safety management practices in the general aviation industry.ResultsOne GAA was analyzed using SCM-GA. The human and organizational causes revealed by SCM-GA were more complete than the causes revealed through the accident report. The identification results of the deficiencies in the subcategories of GA-SMS and the safety culture were more consistent with the requirements in the general aviation laws and regulations than the organizational factors in the accident report. Based on the subcategories of SCM-GA, 41 GAAs that occurred between 1996 and 2010 in China were statistically analyzed and χ² test analyses were performed to estimate the statistical strength of the association between two adjacent subcategories of SCM-GA. The results showed that two adjacent subcategories of SCM-GA were significantly associated. They helped to determine the hidden problems in the accident report based on the path of accident.ConclusionsSCM-GA is an accident analysis tool that can comprehensively analyze the human and organizational deficiencies involved in GAAs. The accident causes revealed by SCM-GA were more consistent with the general aviation safety management practices.Practical applicationsGeneral aviation companies should establish their own GA-SMS and safety culture based on the subcategories developed herein. Using SCM-GA for routine safety inspection and accident investigation will help the management and the staff make effective safety decisions to effectively prevent GAAs.     

公路交通伤亡和行政决策之间关系研究

根据1980年至今的交通伤亡统计资料,笔者对我国公路交通事故现状进行了简要分析,并指出了事故内容统计方法的不足。根据公路交通存在着多部门管理、多部门分配利益,没有部门承担路难责任的问题,提出了业主、司机、警察和法院的职责和关系,把利益和责任联系起来。笔者认为,政府的行政决策非常重要,同时对政府处理路难态度提出自己的观点,对按不同路型不同决策提出建议,对用科技手段解决路难问题提出了加强平等竞争的呼吁。     

10种事故致因模型的对比研究

为了给事故致因模型的研究提供理论支持,为日常管理中事故致因模型的选用提供参考依据,从事故的影响对象、模型的组成和事故发生的路径3个维度对10种事故致因模型进行了对比分析。研究结果表明:事故致因模型对事故影响对象的涵盖越来越全面;事故致因模型的组成日趋全面,且趋于对模型中各模块给出清晰定义以能保证分析结果的准确性;故发生的路径描述由链式向系统网状发展;根据充分性和效率的需求对模块化模型和非模块化的模型进行选择。     

Archetypes for organizational safety

We propose an initial set of six system safety archetypes that model common dynamic organizational behaviours that often lead to accidents. The archetypes help clarify why safety-related decisions do not always result in the desired behaviour, and how independent decisions in different pans of the organization can combine to have a negative, and often unforeseen, impact on safety. Examples from the NASA manned space program and general industry are used to illustrate the archetypes. As accident analysis and investigation tools, the archetypes can be used to develop dynamic models that describe the systemic and organizational factors contributing to accidents. As risk analysis tools, the archetypes can be used to improve understanding of the ways that risk can arise in complex socio-technical systems.     

Introducing roadside hazard severity indicator based on evidential reasoning approach

Safety experts have, in recent years, been attentive to roadside accident severity and occurrence. Hitherto, to prioritize road segment hazardousness, there have been little efforts to quantify a well defined indicator. In this regard, the existing indicators are usually very plain and the overall configuration of roadside is rated by experts with an exact number describing its condition. Hence, the uncertainties which come with the subjective judgments cannot be regarded as of any substance. This research contribution therefore presents a procedure to assess the road safety (roadside safety indicator) by means of the evidential reasoning (ER) approach. The betterment of ER as opposed to the available procedures for roadside safety assessment is that the proposed approach makes allowance for the uncertainties which may arise from individual judgments. Additionally, when there is a dearth of evidence concerning factors which affect roadside hazardousness severity to collate several roadside segments, this procedure will offer the benefits of utilizing the maximum/minimum utility function. With the aid of the drawn indicator, organizations and agencies responsible for ensuring road safety can reach more flexible decisions to set in-place uncertain planning and road segments priorities. This indicator can also be utilized as a variable to include roadside conditions in crash severity prediction models. A field case study has also been carried out in an attempt to follow and validate the proposed approach which is based on the run-off accident history for a sample road segments. The crash data confirm the suggested indicator.     

What-You-Look-For-Is-What-You-Find – The consequences of underlying accident models in eight accident investigation manuals

Accident investigation manuals are influential documents on various levels in a safety management system, and it is therefore important to appraise them in the light of what we currently know – or assume – about the nature of accidents. Investigation manuals necessarily embody or represent an accident model, i.e., a set of assumptions about how accidents happen and what the important factors are. In this paper we examine three aspects of accident investigation as described in a number of investigation manuals. Firstly, we focus on accident models and in particular the assumptions about how different factors interact to cause – or prevent – accidents, i.e., the accident “mechanisms”. Secondly, we focus on the scope in the sense of the factors (or factor domains) that are considered in the models – for instance (hu)man, technology, and organization (MTO). Thirdly, we focus on the system of investigation or the activities that together constitute an accident investigation project/process. We found that the manuals all used complex linear models. The factors considered were in general (hu)man, technology, organization, and information. The causes found during an investigation reflect the assumptions of the accident model, following the ‘What-You-Look-For-Is-What-You-Find’ or WYLFIWYF principle. The identified causes typically became specific problems to be fixed during an implementation of solutions. This follows what can be called ‘What-You-Find-Is-What-You-Fix’ or WYFIWYF principle.     

Reference criteria for the identification of accident scenarios in the framework of land use planning

Land use planning (LUP) around industrial sites at risk of major accidents requires the application of sound approaches in the selection of credible accident scenarios. In fact, the ‘technical’ phase of LUP is based on the identification and assessment of relevant accident scenarios. An improper choice of scenarios may critically affect both the ‘technical’ phase of risk assessment and the following ‘policy’ phase concerning decision making on land-use restrictions and/or licensing. The present study introduces a procedure aimed at the systematic identification of reference accident scenarios to be used in the gathering of technical data on potential major accidents, which is a necessary step for LUP around Seveso sites. Possible accident scenarios are generated by an improved version of the MIMAH methodology (Methodology for the Identification of Major Accident Hazards). The accident scenarios are then assessed for LUP relevance considering severity, frequency and time scale criteria. The influence of prevention and mitigation barriers is also taken into account. Two applications are used to demonstrate the proposed procedure. In both case-studies, the proposed methodology proved successful in producing consistent sets of reference scenarios.     

Transferability of accident prediction models

Accident prediction models, the vast majority of which are negative binomial regression models, are of considerable importance to highway agencies since they can be used to conduct many traffic safety studies. However, not every agency possesses sufficient accident statistics that enable it to develop reliable models of its own. This problem gives rise to interest in the transferability of accident prediction models in time and space. It would save time, effort, and money if accident prediction models developed for one region in one period of time could be applied in different time periods and regions to produce reliable safety studies.This paper presents methods for recalibrating negative binomial accident models before transferring them for use in different time periods and regions of space. The paper emphasizes that the recalibration of the shape parameter of a transferred model using local data is absolutely necessary. It explains that it is also desirable to recalibrate the constant term of the transferred model in order to allow the model to better suit local conditions. A moment method is presented for recalibrating the shape parameter of a transferred model when its constant term is not recalibrated. However, a maximum likelihood method is presented for recalibrating both the shape parameter and the constant term of the transferred model and is shown to be superior to the recalibration methods existing in the traffic safety literature.     

Investigation of key factors for accident severity at railroad grade crossings by using a logit model

Although several studies have used logit or probit models and their variants to fit data of accident severity on roadway segments, few have investigated accident severity at a railroad grade crossing (RGC). Compared to accident risk analysis in terms of accident frequency and severity of a highway system, investigation of the factors contributing to traffic accidents at an RGC may be more complicated because of additional highway–railway interactions. Because the proportional odds assumption was violated while fitting cumulative logit modeled by the proportional odds models with stepwise variable selection to ordinal accident severity data collected at 592 RGCs in Taiwan as suggested by Strokes et al. [Strokes, M.E., Davis, C.S., Koch, G.G., 2000. Categorical Data Analysis Using the SAS System, second ed. SAS Institute, Inc., Cary, NC, p. 249], a generalized logit model with stepwise variable selection was used instead to identify explanatory variables (factors or covariates) that were significantly associated with the severity of collisions. Hence, the fitted model was used to predict the level of accident severity, given a set of values in the explanatory variables. Number of daily trains, highway separation, number of daily trucks, obstacle detection device, and approaching crossing markings significantly affected levels of accident severity at an RGC (p-value = 0.0009, 0.0008, 0.0112, 0.0017, and 0.0003, respectively). Finally, marginal effect analysis on the number of daily trains and law enforcement camera was conducted to evaluate the effect of the number of daily trains and presence of a law enforcement camera on the potential accident severity.     

Intervention strategies for accident-involved drivers: An experimental evaluation of current California policy and alternatives

This project was designed to evaluate current and alternative strategies for selecting and treating accident-involved drivers in California. The current accident-based selection criteria were contrasted with an expanded selection strategy that included convictions as well as accidents as criteria for selection. The standard diagnostic reexamination treatment was compared to two alternative behavior modification treatments—an accident avoidance session and a mailed educational pamphlet/self-administered test. Interaction between treatments and selection criteria revealed that accident-involved drivers with minimum conviction histories were amenable to treatment intervention, while accident-involved drivers with more extensive conviction histories were not. Treatment effects were statistically significant (p < .02), with each of the treatment groups having posttreatment accident means in excess of 20% lower than comparable control groups.     

Optimum placement of sea rescue resources

In countries with autonomous regional governments, the positioning of the national sea rescue resources is often a permanent source of friction between the national and regional authorities. This friction usually resurges after any heavily publicised accident. However, the process of planning sea rescue resources and their distribution in the various locations should be carried out according to scientific criteria.The aim of the present work is to build a tool which allows sea rescue resources to be assigned objectively. To this end, we formalise a general methodology based on gravitational models which allows us to define individual and zonal distribution models. Also, a practical application of the zonal model is performed, assigning ‘sea rescue boats’ to a segment of the coast where there are three ports.     

Dynamic modeling and bifurcation analysis for the methyl isocyanate hydrolysis reaction

The aim of this work was to characterize the methyl isocyanate hydrolysis reaction and to identify its operational criteria. The parametric sensitivity and dynamic stability methodologies were performed at the Bhopal disaster circumstances, over the relevant operating parameter space. Stable and unstable conditions, bifurcations points, turning points and oscillatory behavior were determined. The combined methodology give useful guidance on the operative conditions selection and the appropriate strategy to overcome hazardous situations. The obtained results demonstrated high sensitivity to small perturbations (thermal runaway) and prevalent oscillatory behavior. Moreover, the following critical parameters for the studied dynamic system were defined: the inverse residence time of 1.5700103 and the heat transfer coefficient of 752.394.     

Time series analysis of coal mine accident experience

This study investigates several forecasting techniques that can be useful to mine safety managers for studying mine accident rate behavior. Three time series models were studied for extrapolation of accident rates. These models are applied to historical accident incidence data from a coal mine. Further, a method is presented for evaluating the three models for the selection of an appropriate model. For this particular mine application, it is concluded that the more complex Box-Jenkins ARMA model as well as first order autoregressive model do not give better results than the simple exponential smoothing model. However, when the random variations or autocorrelations in the accident experience rates between periods are different, the models may predict differently. As such, specific models must be developed for each mine on the basis of statistical analysis of the mine accident experience data over time. Moreover, the importance of incorporating human judgement to interpret the results of statistical forecasting cannot be overemphasized. Integration of policy or operating changes, which may impact mine safety performance, with statistical forecasting techniques is essential to arrive at a realistic prediction of future performance.     

A meta‐analytic comparison of self‐reported and other‐reported organizational citizenship behavior

Given the common use of self‐ratings and other‐ratings (e.g., supervisor or coworker) of organizational citizenship behavior (OCB), the purpose of this meta‐analysis was to evaluate the extent to which these rating sources provide comparable information. The current study's results provided three important lines of evidence supporting the use and construct‐related validity of self‐rated OCB. The meta‐analysis of mean differences demonstrated that the mean difference in OCB ratings is actually quite small between self‐ and other‐raters. Importantly, the difference between self‐ and other‐raters was influenced by neither the response scale (i.e., agreement vs. frequency) nor the use of antithetical/reverse‐worded items on OCB scales. The meta‐analysis of correlations showed that self‐ and other‐ratings are moderately correlated but that self–other convergence is higher when antithetical items are not used and when agreement response scales are used. In addition, self‐ratings and supervisor‐ratings showed significantly more convergence than self‐ratings and coworker‐ratings. Finally, an evaluation of self‐rated and other‐rated OCB nomological networks showed that although self‐rated and other‐rated OCBs have similar patterns of relationships with common correlates, other‐rated OCB generally contributed negligible incremental variance to correlates and only contributed appreciable incremental variance to other‐rated behavioral variables (e.g., task performance and counterproductive work behavior). Implications and future research directions are discussed, particularly regarding the need to establish a nomological network for other‐rated OCB. Copyright © 2013 John Wiley & Sons, Ltd.     

基于BERT-BILSTM-CRF模型的电力行业事故文本智能分析*

为解决电力行业事故报告文本较长、语义复杂,难以进行有效文本识别问题,提出1种以BERT作为底层的预训练模型,并设计1种双重注意力机制编码器,结合BILSTM-CRF深度挖掘事故文本语义特征,从而实现文本智能分析。首先构建电力词典,通过对BERT预训练,进行BIO标注,然后引入BILSTM-CRF模型实现对文本标签智能分类,最后将该模型与现行其他4种深度学习模型进行对比。研究结果表明:该模型智能识别精确率、召回率及F1值(查准率)均达到约97%,较其他4种模型中效果最好的模型分别提高0.02,0.03,0.02。研究结果可为电力行业事故报告文本分析提供1种新思路。