A probabilistic approach for determining the control mode in CREAM
Introduction
Since Dougherty's guest editorial [1], many Human Reliability Analysis (HRA) methods have been developed to overcome the shortcomings of 1st generation HRA methods. They are called 2nd generation HRA methods. Among them, ATHEANA (A Technique Human Error ANAlysis) [2] and CREAM (Cognitive Reliability and Error Analysis Method) [3] are the best known methods.
In many 1st generation HRA methods such as THERP (Technique for Human Error Rate Prediction) [4], ASEP (Accident Sequence Evaluation Program) [5] and HCR (Human Cognition Reliability) [6], the basic assumption is that because humans have inherent deficiencies, humans naturally fail to perform tasks, just like mechanical or electrical components. Therefore, the probability of human failure can be assigned based on the characteristics of the task that human has to perform. In this sense, the term Human Error Probability (HEP) is defined. The HEP is modified by the factors representing the effects of the environment. Usually those factors are called Performance Shaping Factors (PSFs) or Performance Influencing Factors (PIFs). In 1st generation HRA methods, the characteristics of a task, which is represented by HEPs, are regarded as the major factors and the environment, which is represented by PSFs and PIFs, is considered as a minor factor, in estimating the probability of human failure.
As time goes on, however, it was found that the effect or importance of the environment is larger than that of the characteristics of a task [7]. In other words, the environment or the context is considered as the major factor and the characteristics of a task are considered as minor factors. This led the change in the focus of human failure, and became the basic principle of 2nd generation HRA methods.
If the environment or the context is the major factor for the estimation of human failure probability, the relation between the context and the human failure probability should be specified. In CREAM, more specifically in Contextual Control Model (COCOM) [8] which is the human cognition model used in CREAM, it is assumed that the most important clue to estimating human performance or human failure probability is the degree of control that human operators have over the situation (or context). And, it is also assumed that the degree of control can be determined by the context under which human operators perform their tasks. As a whole, the degree of control is the core concept that defines the relation from the context to human failure probability.
Even though the degree of control is a continuous value, the division of the degree of control into four categories is suggested in CREAM. Each of the four categories is called a control mode. The four control modes are scrambled control, opportunistic control, tactical control and strategic control, by the ascending order of the degree of control. The estimated human failure probability interval for some action in each control mode is also provided in CREAM.
In CREAM, nine Common Performance Conditions (CPCs) are defined as a minimum number of factors that are important to describe the context (the 9 CPCs can be found in Table 1). As shown in Table 1, each CPC has finite number of levels. For example, the CPC adequacy of organization has four different levels, very efficient, efficient, inefficient and deficient. The level of each CPC, i.e. the state of each CPC, is assessed by analysts. After the assessment of the levels of CPCs, the control mode is determined. CREAM provides a method for determining the control mode from the assessed levels of CPCs. In the following section, we will briefly describe the existing method for determining the control mode, and then explain the advantages of a probabilistic approach. After that, the procedure for deriving the probabilistic model for determining the control mode will be described in detail.
Section snippets
The existing method (Deterministic)
To illustrate the existing method for determining the control mode from the assessed levels of CPCs, we show how the existing method can be applied to an example context which is shown in Table 1. As shown in Table 1, even though different levels are used for different CPCs, i.e. different sets of states are used for different CPCs, the expected effect on performance reliability is described using one set of levels, {improved, not significant, reduced}. From the expected effects on performance
Examples
The first example is for the case of determining the control mode when the levels of 9 CPCs are given deterministically, to show that the Bayesian network developed in this paper produces equivalent results with the existing method. Fig. 1 shows the Bayesian network and the calculation result for the context described in Table 1. We used a non-commercial software tool, MSBNx, which can be downloaded at http://research.microsoft.com/adapt/MSBNx/. As shown in Fig. 1, the control mode is
Conclusions
The control mode is the core concept for the prediction of human performance in CREAM. Because the existing method has several limitations in considering the uncertainties associated with the specification of CPCs, we proposed a new method, which is based on the probabilistic approach. The probabilistic approach is found to have several advantages over the existing method.
From the fact that any discrete function can be expressed in a Bayesian network, which is a probabilistic model, the
Acknowledgements
This work is partly supported by Korean National Research Laboratory (NRL) program.
References (9)
Human reliability analysis—where shouldst thou turn?
Reliab Eng Syst Safety
(1990)- NUREG-1624. Technical basis and implementation guidelines for a technique for human event analysis (ATHEANA), Rev.1,...
Cognitive reliability and error analysis method
(1998)- Swain AD, Guttman HE. Handbook of human reliability analysis with emphasis on nuclear power plant applications,...
Cited by (100)
A novel integrated methodology for human reliability assessment in hydrogen fuelling stations
2023, International Journal of Hydrogen EnergyAn inter-method comparison of four Human Reliability Assessment models
2022, Applied ErgonomicsPredicting human reliability based on probabilistic mission completion time using Bayesian Network
2022, Reliability Engineering and System Safety