Human error probability estimation by coupling simulator data and deterministic analysis

https://doi.org/10.1016/j.pnucene.2015.01.008Get rights and content

Highlights

  • Uncertainty in operator action time and allowed time is combined.

  • Probability distribution function for operator time is determined.

  • Probability distribution function for allowed time is determined.

  • HEP is predicted from these two distributions.

Abstract

Operator error in diagnosis and execution of task have significant impact on Nuclear Power Plant (NPP) safety. These human errors are classified as mistakes (rule base and knowledge based errors), slip (skill based) and lapses (skill based). Depending on the time of occurrence, human errors have been categorized as i) Category ‘A’ (Pre-Initiators): actions during routine maintenance and testing wherein errors can cause equipment malfunction ii) Category ‘B’ (Initiators): actions contributing to initiating events or plant transients iii) Category ‘C’ (Post-Initiators): actions involved in operator response to an accident. There have been accidents in NPPs because of human error in an operator's diagnosis and execution of an event. These underline the need to appropriately estimate HEP in risk analysis. There are several methods that are being practiced in Probabilistic Safety Assessment (PSA) studies for quantification of human error probability. However, there is no consensus on a single method that should be used. In this paper a method for estimating HEP is proposed which is based on simulator data for a particular accident scenario. For accident scenarios, the data from real NPP control room is very sparsely available. In the absence of real data, simulator based data can be used. Simulator data is expected to provide a glimpse of probable human behavior in real accident situation even though simulator data is not a substitute for real data. The proposed methodology considers the variation in crew performance time in simulator exercise and in available time from deterministic analysis, and couples them through their respective probability distributions to obtain HEP. The emphasis is on suitability of the methodology rather than particulars of the cited example.

Introduction

The human errors in the performance of desirable diagnosis and actions during an accident situation have significant contribution to the risk. The actual estimates of the fractional contributions of human error to system failures have varying quantitative values. However, many analysts have indicated that the fraction could be as high as 50% for full-power operations (IAEA TECDOC-565, 1990) and as high as 70% for low power and shutdown state of Nuclear Power Plant (Himanen, 1995). The fact that the contribution of human error could be high in over all risk, it is important that Human Error Probability (HEP), is correctly estimated for the purpose of Probability Safety Assessment (PSA). In order to accomplish this requirement, it is necessary to select a suitable method for estimation of HEP.

The human errors have been categorized as (IAEA 50-P-10, 1995): (i) Category ‘A’ (Pre-Initiators) – These consist of actions associated with maintenance and testing which degrade system availability. They may cause failure of a component or component group or may leave components in an inoperable condition. Some examples of pre-initiators are mis-calibration of sensors, valve misalignment, incorrect part fitting during maintenance and working on wrong component. (ii) Category ‘B’ (Initiators) – These are actions contributing to initiating events or plant transients. They are implicit in selection of initiating events for PSA. (iii) Category ‘C’ (Post-Initiators) – These are the actions involved in operator response to an accident. The post-initiators are generally classified into procedural safety actions, aggravating actions and recovery actions. Category ‘C’ actions have always been at the center of HEP because they are critical for NPP safety.

The human error is classified under three types (i) Mistakes – The action is intended to be performed as planned but wrong course is taken thinking it to be correct. Mistakes could be rule and knowledge based. Example of rule based mistake is misapplication of a correct procedure or correct application of a badly/wrongly written procedure. The knowledge based mistake could be due to non existence of a procedure for an unusual situation and reliance on gathered experience and knowledge over time (ii) Slip – These are associated with familiar task which do not require much of conscious thinking. They lead to commission of errors. (iii) Lapses–Lapses are also linked to tasks not requiring conscious effort. They lead to omission of errors.

The pioneering Reactor Safety Study or popularly called the WASH-1400 (WASH, 1975) was the first to address the issue of Human Reliability contribution to system unavailability. The field of Human Reliability Assessment has gone through several stages of development and detailing. In the last two decades several methods have been proposed (Health and Safety Executive, 2009), and used in the nuclear industry. The HRA methods are broadly classified into three categories. These categories are i) Task Related, ii) Time Related, iii) Context Related. The task related and time related categories constitute the first generation methods while the context related category constitutes the second generation methods.

There have been few bench marking exercises for HRE estimation. In the study (Poucet, 1989), the HRA methods THERP, SLIM, HCR, HEART, Technica Empirica Stia Errori Operatori (TESEO), Absolute Probability Judgment (APJ) and Maintenance Personnel Performance Simulation (MAPPS) were compared. In the paper Boring et al. (2010), some of the observations have been mentioned. There was considerable variability in the estimates obtained from different methods (many order of magnitude difference). The inter method reliability was low. Also, the reliability of the results obtained by different experts from one method was also low. In another paper (Kirwan, 1997), empirical validation of three HRA methods, namely THERP, HEART and Justification of Human Error Data Information (JHEDI) was carried out. The paper Boring et al. (2010) mentions the lesson learnt in this study. It is pointed out that there were difficulties in consistently modeling error of commission in HEART and JHEDI, slips in HEART, diagnostic task in THERP and human machine interface task in THERP. These shortcomings bring home the point that no HRA method is comprehensive in its coverage of human errors and that each method represents strength and weaknesses in terms of its coverage and quantification. For the Cognitive Reliability Error Analysis Method (CREAM) it was noted (Kirwan, 1988) that “these approaches are potentially of most interest to psychologists and others who want to predict the more sophisticated error forms associated with misconceptions, misdiagnosis, etc. They attempt to explore the error forms arising from ‘higher-level’ cognitive behaviours”. There is not much literature which suggests extensive use of CREAM in NPPs. ATHEANA methodology is cumbersome, guidance is complex, costly to implement, uses expert judgment for quantification and hence may be less reliable (HSE, 2009). The document (HSE, 2009), gives a brief summary of 17 HRA methods along with their advantages and disadvantages.

Section snippets

HRA methods – a brief overview

The HRA methods Technique for Human Error Rate Prediction (THERP, Swain and Guttman, 1983) and ASEP (Swain, 1987) are the foremost of the task analysis based methods. Both these methods require the analyst to draw an Operator Action Tree (OAT) based on a detailed analysis of the task to be performed. The critical assessment of THERP and ASEP methods brings out the weakness of interpreting crew performance as a sequence of individual tasks. In addition, the use of look up tables for human error

Proposed methodology

Event progression following an initiating event can be characterized as a series of functional failures or unavailable states of relevant mitigating systems or the occurrence of new events. Estimating human error probability is a highly complicated task since it involves a huge number of variables which are internal and external to the working environment in an NPP. The internal factors to the working environment include the physical environment, ergonomics (Carvalho, 2006), system design,

Probability distribution functions and method of least squares

The raw data on Tc and Ta needs to be fit to the most probable distribution. There are a number of probability distributions that could be used to check for the data fitting. In this study, the method of least squares has been used for data fitting. This is described below for four probability distributions.

HEP estimation

A dummy data set for crew timing, Tc, and time available by deterministic calculation, Ta, is used for demonstrating the HEP estimation by coupling of these data. The system time Ta is obtained as the time available to reach a reactor parameter to an unacceptable value (e.g. fuel clad temperature). The variations in Tc and Ta of the dummy data for a hypothetical event are shown in Table 3.

Discussion

The estimation of HEP was carried out using point estimate, probability distributions for Ta and Tc, and HCR model. The point estimate values for HEP gives an unreasonable and impractical result because there are inherent differences in action time among operators. The HCR methodology also, has given very high value for HEP. HCR assumes that the data follows a Weibull type distribution. The assumption may not be correct as can be seen from the data which fits normal distribution with high

Conclusion

The review of research work on HEP estimation shows that there is no universal methodology to estimate HEP. The question what is the correct HEP is only true in a statistical sense. The proposed method emphasizes that accident scenario specific simulator data generated from many exercises is suitable for reliability analysis. This is more realistic than using the generic human error data that could have been generated for a different accident scenario and in a different regional setting. The

References (26)

  • D. Gertman et al.

    The SPAR-H Human Reliability Analysis Method

    (2004)
  • G.W. Hannaman et al.

    Human Cognitive Reliability Model for PRA Analysis

    (1984)
  • Health and Safety Executive (HSE)

    Review of Human Reliability Assessment Methods

    (2009)
  • Cited by (12)

    • Human reliability analysis studies from simulator experiments using Bayesian inference

      2023, Reliability Engineering and System Safety
      Citation Excerpt :

      If the operator performance data is required for an advanced (first of its kind of) reactor, the available data of other plants may not be applicable. In such a situation, simulators can be used to generate the HRA data [14]. The quality/ applicability of the HRA data generated in case of simulator would depend on the degree of replication of the plant's control room.

    • Human factors of flight training and simulation

      2022, Human Factors in Aviation and Aerospace, Third Edition
    • Evaluation of risk impact of completion time changes combining PSA and DSA model insight and human reliability analysis

      2018, Reliability Engineering and System Safety
      Citation Excerpt :

      Several studies have focused on the benefits of using DSA and PSA together to improve the accuracy time windows of operator actions, i.e. the time interval in which operators have to perform an action to make the plant safe. Calculating Human Error Probability (HEP) can be done by HRA techniques [13–16], while Thermal Hydraulic (TH) simulations using best estimate codes could evaluate the appropriateness of the accident scenarios pre-established by PSA models, identifying and characterizing unknown accident sequences and success criteria [17]. Starting from the International Atomic Energy Agency's (IAEA) structured framework [18], the combined results of both approaches provide an input on integrated risk-informed decision making to ensure nuclear reactor safety [19–22].

    • A classification scheme of erroneous behaviors for human error probability estimations based on simulator data

      2017, Reliability Engineering and System Safety
      Citation Excerpt :

      However, the failures associated with a timing issue for each task were not counted in this study. Because the data collected from the simulators includes the information of human performance times on given tasks, it is expected to estimate a HEP by comparing the performance time distribution with a thermal hydraulic criteria as shown in [46,47]. In this study, we proposed a classification scheme of erroneous behaviors in terms of HEP estimations from simulator data.

    • A hybrid human error probability determination approach: The case of cargo loading operation in oil/chemical tanker ship

      2016, Journal of Loss Prevention in the Process Industries
      Citation Excerpt :

      While the AHP technique quantifies the subjective judgement of experts and ensures expert elicitation verification, HEART provides quantification of HEP for specific task that is being assessed. Since human error and recovery are key attributes of a risk management in various domains, there are a wide range of approaches used to account for the human error contribution to risk management (Hameed et al., 2016; Akyuz, 2015, 2016; Prasad and Gaikwad, 2015; Sun et al., 2012; Arslan, 2009; Reniers, 2009). Most of studies discussed role of human error in different industries and addressed this issue to fatigue and insufficient training of operator.

    View all citing articles on Scopus
    View full text