Human error probabilities

The analysis of many accidents has led to the appreciation that multiple equipment failures and process deviations combined with faulty human decisions and actions are often involved. Safety assessments, therefore, are not complete unless the interactions between equipment failures and human actions are considered. Since human behaviour is complex, and does not lend itself immediately to relatively straightforward reliability models, it is suggested that the following classifications of human interactions (that typically group all activities) need to be considered (Mahn et al. (1995))  

These classifications of human interactions can be related to a simple error classification system consisting of three categories (1) slips, (2) non-response, and (3) mistakes. This classification scheme can then be used to qualitatively incorporate human errors in accident scenarios. Table 9.1 provides generic human error probabilities for use in accident scenario 

The development of a generic set of human error probabilities is extremely difficult since there is a strong correlation on the actual person performing the task, the complexity of the task, the time required for task completion, and the training level of the person performing the task. Additionally, a worker may perform any specific task differently depending on the level of alertness due to fatigue or other factors. 

---

Human error probabilities can also be estimated using methodologies and techniques originally developed in the nuclear industry. A number of different models are available (Swain, Comparative Evaluation of Methods for Human Reliability Analysis, GRS Project RS 688, 1988). This estimation process should be done with great care, as many factors can affect the reliability of the estimates. Methodologies using expert opinion to obtain failure rate and probability estimates have also been used where there is sparse or inappropriate data. 

Phase three of a typical HRA begins with developing human error probabilities that can he applied to the selected model. In some cases, a set of nominal human errors can be derived 1mm plant data, however, due to the sparseness and low confidence of these data industry generic information may be used. Chapter 20 of NUREG/CR-1278 includes a typical set of. such data. 

The human error probabilities estimated for a given task can now be modified to reflect the actual performance situation. For example, if the labeling scheme at a particular plant is very poor, the probability should be increased towards an upper bound. If the tagging control system at a plant is particularly good, the probability for certain errors should be decreased toward a lower bound. 

In any given situation, there may be different levels of dependence between an operator s performance on one task and on another because of the characteristics of the tasks theraseb e.s. or because of the manner in which the operator was cued to perform the tasks. Dependence levels between the performances of two (or more) operators also may differ. The analyses should account for dependency in human-error probabilities. In addition, each sequence may have a set of human recovery actions that if successfully performed will terminate or reduce the consequences of the sequence. This information, coupled with a knowledge of the system success criteria leads to the development of human success and failure probabilities which are input to the quantification of the fault iices or event trees. With this last step, the HRA is integrated into the PSA, and Pl. ise 4 is complete. 

List of the generic human error probabilities used to determine a base error rate for each human error considered, and... 

Table 4.5-11 Sample o/NUCLARR Human Error Probability Data iNUREG/CR-4639f...

The development of the HRA event tree is one of the most critical parts of the quantification of human error probabilities. If the task analysis lists the possible human error events in the order of ihcir potential occurrence, the transfer of this information to the HRA event tree is fadlitutcd. Each potential eiTor and success is represented as a binary branch on the HRA event tiec. with subsequent errors and successes following directly from the immediately preceding ones. Cure should be taken not to omit the errors that are not included in the task analysis table but might affect the probabilities listed in the table. For example, administrative control errors that affect a task being performed may not appear in the task analysis table but must be included in the HRA event tree. 

Human reliability [lata NJUREG/CR-1278 was supplemented by judgment of system analysts and plant personnel. Human error probabilities were developed from NUREG/CR-12 8, human action time windows from system analysis and some recovery limes from analysis of plant specific experience. Data sources were WASH-1400 HEPs,Fullwood and Gilbert assessment ot I S power reactor Bxp., NUREG/ CR -127K. and selected acro ptice li.it.j... 

Voska, K.J. and J.N. O Brien, Human Error Probability Estimation Using Licensee Event Reports, BNL, July 1984. 

V) Reece, W. J. et al., Nuclear Computerized Library for Assessing Reactor Reliability (NUCLARR), Part. 2 Human Error Probability Data (HEP), 1994. 

Chapter 5, Quantitative and Qualitative Prediction of Human Error in Safety Assessments, describes a systematic process for identifying and assessing the risks from human error, together with techniques for quantifying human error probabilities. 

Chapter 4 focuses on techniques which are applied to a new or existing system to optimize human performance or qualitatively predict errors. Chapter 5 shows how these teclmiques are applied to risk assessment, and also describes other techniques for the quantification of human error probabilities. Chapters 6 and 7 provide an overview of techniques for analyzing the underlying causes of incidents and accidents that have already occurred. 

In numerical terms, the probability of each failure state is given by the following expressions (where SP is the success probability and HEP the human error probability at each node) ... 

The decomposition approach is used, it is necessary to represent the way in which the various task elements and other possible failures are combined to give the failure probability of the task as a whole. Generally, the most common form of representation is the event tree (see Section 5.7). This is the basis for THERP, which will be described in the next section. Fault trees are only used when discrete human error probabilities are combined with hardware failure probabiliHes in applications such as CPQRA (see Figure 5.2). 

History and Technical Basis. The influence diagram approach (IDA) (also known as the sociotechnical approach to human reliability (STAHR) (see Phillips et al., 1990) is a technique that is used to evaluate human error probabilities as a... 

Quantitative human reliability data collection systems for generating human error probabilities for use in quantitative risk assessment. 

There is considerable interest in developing a database on human error probabilities for use in chemical process quantitative risk assessment (CPQRA). Nevertheless, there have been very few attempts to develop such a database for the CPI compared, for example, with the nuclear industry. Some of the reasons for this are obvious. The nuclear industry is much more highly integrated than the CPI, with a much greater similarity of plant equipment... 

Embrey, D. E., Kirwan, B., Rea, K., Humphreys, P., Rosa, E. A. (1984). SLIM-MAUD. An approach to Assessing Human Error Probabilities Using Structured Expert judgment Vols. I and II. Washington, DC NUREG/CR—3518 US Nuclear Regulatory Commission. 

Human Error Probability The probability that an error will occur during the performance of a particular job or task within a defined time period. Alternative definition The probability that the human operator will fail to provide the required system function within the required time. 

About 1,000 miscellaneous failure rates, event rates, and probabilities. There is some treatment of human error probability... 

After repair, the component is returned to the working state. Minimal cut set analysis is a mathematical technique for developing and providing probability estimates for the combinations of basic component failures and/or human error probabilities, which are necessary and sufficient to result in the occurrence of the top event. 

Human error after maintenance HUM Q = HEPfmc Ts fmc = Corrective maintenance frequency HEP = Human error probability... 

In direct numerical estimation, decision makers are asked to give a numerical estimate of how likely they think the event is to happen. These estimates can be probabilities, odds, log odds, or words (Winterfeldt and Edwards 1986). Winterfeldt and Edwards argue that log odds have certain advantages over the other measures. Gertman and Blackman (1994) note that log odds are normally used in risk assessment for nuclear power applications because human error probabilities (HEPs) vary greatly in value. HEPs between 1 and 0.00001 are typical. 

SLIM-MAUD (Embrey 1984) implements a related approach in which expert ratings are used to estimate human error probabilities (HEPs) in various environments. The experts first rate a set of tasks in terms of performance-shaping factors (PSFs) that are present. Tasks with known HEPs are used as upper and lower anchor values. The experts also judge the importance of individual PSFs. A subjective likelihood index (SLI) is then calculated for each task in terms of the PSFs. A logarithmic relationship is assumed between the HEP and SLI, allowing calculation of the human error probability for task j (HEPj) from the subjective likelihood index assigned to task j (SLIj). More specifically ... 

After the task is specified, human and machine performance models can be applied to estimate task performance. The MHP and keystroke-level performance model can provide task performance estimates in terms of task completion time. THERP can be used to estimate human error probabilities for each task and task sequence. The ERM approach can be used to estimate performance along any required dimension and to compare required with available resources along any required dimension as long as the human performance data are available. The results of the ERM assessment would identify stress levels on capacities (e.g., resources stressed too long or beyond maximum capacity). These results indicate limiting factors to successful task performance. Limiting factors can be identified at elemental or intermediate performance resource levels. As such, the ERM represents a more comprehensive and internally consistent model than the others. It is more comprehensive in that it can be used to model any performance dimension. It is... 

Planning errors cause wrong designs and construction flaws and lead to wrong or insufficient instructions in the operating manual. Planning errors stem, for example, from mutual dependencies which have not been identified or sufficiently accounted for as the dependence of human error probabilities on environmental influences or impairment of components due to changes in environmental conditions caused by an accident. 

---