1 Quantifying Hardware Failure
Hardware failure or component failure refers to the unavailability
of the respectively components when they are required to perform its function
in the system. To quantify a component’s failure rate, we can refer
to the inventory data sheet obtainable from the manufacturer of the respective
items. Of course, expert judgement can play a part in determining
the final probability of failure. In this case, we may need
Bayesian methods to provide us the estimates for the failure of the component.
Figure 1 shows a information flow sheet in a reliability data system.
Figure 1 : Information flow sheet in a reliability data system.
(Ernest J. Henley and Hiromitou Kumanto, 1992)
Failure rates for various instruments used in porcess industries can
be found in literature [3]. It also provide references to which further
data could be obtained. Other sources of data are documented in [1,
7, 5]. Of course, one can also inquire for such data sources using
the Internet by contacting Reliability Analysis Centre[51].
BACK to TOP
BACK to HOME
2 Human Performance Error
Before discussing the methodologies available for the quantification
of human failures, it is necessary to have a taxonomy of the various human
error. This is important as it eliminates imaginative but unsupportable
human failures in the logic structure of the HRA or fault tree.
2.1 Qualitative Analysis
of Human Failures
James Reason(1993)[32] classified human failure in the productive
element framework into two broad groupings, termed types and tokens.
Types are general classes of organisational and managerial failures.
Tokens are more specific failures relating to individuals at the human-system
interface. Both classes of failures are then sub-categorised as in
figure 2.
Figure 2 : Sub-categories for types and tokens
(James Reason, 1993).
The understanding of the different types of human failure is essential
to the selection of the right HRA methods to quantify them.
Of course, there exists other taxonomy of human failures such as that of
Rasmussen[35], which provides descriptions regarding the skill-based, rule-based,
or knowledge-based behaviour underlying the execution of tasks.
BACK to TOP
BACK to HOME
2.2 Methods for Quantification
of Human Failures
Most method for estimating human reliability were used in the nuclear
power plant. Such methods include confusion matrix[5, 12], expert
estimation[5, 12], time reliability curve(TRC)[5, 12, 33, 34], maintenance
personnel performance simulation (MAPPS)[5, 12], success likelihood index
method-multi-attribute utility decomposition(SLIM-MAUD)[5, 12, 33], sociotechnical
assessment of human reliability[12], technique of human error rate prediction(THERP)[5,
12, 33, 50], Sandia recovery model(SRM)[12] , INTENT[36] ,and operator
reliability calculation and assessment (ORCA)[12, 33]. Of the methodologies
given, most deals with misdiagnosis or non response errors and time dependent
probability estimates. In the discussions that follows, the most
commonly used technique THERP, utilising generic HEPs from various industries,
and SLIM-MAUD, using importance weightings from experts will be presented.
2.2.1 Technique for
human error rate prediction(THERP)
This method provides mechanism for modeling as well as quantifying
the human error. It start off with a task analysis that describes
the tasks to be performed by the crew, maintainer or operator. Together
with the task descriptions, performance shaping factors(PSF) such as stress
and time available, are collected to modify probabilities. The task
analysis is then graphically represented in HRA event trees. The
human error probabilities(HEPs) for the activities of the task or the branches
are read and/or modified from the THERP tables as shown in Table 5-8 to
5-37 in Gertman and Blackman(1994). Details on the construction of
HRA event trees and also, the cognitive event tree (COGENT) to represent
cognitive activities and errors associated with human performance were
also given in the book. Gertman and Blackman also provides a summary
of the steps to approach THERP which was adapted from the NUREG/CR-1278
(Swain and Guttman 1983). On the other hand, a step by step approach
for quantifying the THERP is documented in literature [33], while the book
[50] gives a brief description of the method.
BACK to TOP
BACK to HOME
2.2.2 Success likelihood index
method-multiattribute utility decomposition(SLIM-MAUD)
The SLIM-MAUD is based on the assumption that the failure probability
associated with task performance is based on a combination of PSFs that
include the characteristics of the individual, the environment, and the
task. It further assumes that experts can estimates these failure
rates or provide anchor values to estimate them. Refer to Gertman and Blackman(1994)[12]
for a description on the steps to take to perform SLIM. It also discussed
two enchanced methods for the SLIM. Dougherty and Fragola[33] also
provides the mathematics and an example for calculating SLI. For
detail description on the SLIM approach, refer to NUREG/CR-3518 Volumes
I and II. On the other hand, literature [38] provides an empirical
evaluation of THERP, SLIM and ranking to estimate HEPs, through the use
of a simulated manufacturing environment under varying task conditions.
2.2.3 INTENT
This method use SNEAK[36] analysis and HSYS[12, 36] together with two
data sources Nuclear Computerised Library for Assessing Reactor Reliability
(NUCLARR)[12, 36] and licensee event reports(LERs)[12, 36] to identify
a generic list of twenty potential errors which may be manifested as erroneous
acts. From experts judgement, the corresponding human error probabilities(HEPs)
in lower and upper bounds were generated. Normalised importance weights
were also computed for each of the 11 performance shaping factors.
The specific ratings for the PSFs together with these generic weights were
then used to compute a composite PSF score, to which is mapped onto an
HEP distribution. The HEP for the decision-based error in a specific
plant is thus obtained. A point to note here is that the HEPs then
obtained are based on expert judgement and not empirical based. These
values should be used judiciously and replaced when operation data are
available.
BACK to TOP
BACK to HOME
3 Management
and Organisational factors in PRA
Apostolakis et al.(1994)[37, 38] reported the development of a methodology
to incorporate the influence of management and organisational factors in
the safety analysis of nuclear power plant. The first paper discussed
the qualitative method, work process analysis model (WPAM-1). Basically
a task analysis is performed on the work process to which the tasks involved,
actions and the defenses in the task, and their failure modes are investigated.
The next step was to define the organisational factors matrix for each
key work process. The matrix specifies task specific organisational
factors by collecting related procedures/documents on the work process
of interest, by conducting interviews with plant interviews with plant
personnel, and by collecting information on plant operating experience.
The organisational factors influencing each task in the given work process
are then ranked according to their importance through the use of the analytical
hierarchy process(AHP). The output of the WPAM-I will then be used
as inputs to WPAM-II.
WPAM-II is used to modify minimal cut set frequencies to include organisational
dependencies among the probabilistic safety analysis(PSA) parameters, ie.
candidate parameter group. Candidate parameter groups(CPGs) are the
group of parameters whose numerical values might change due to the influence
of organisational factor[38]. The minimum cut sets(MCS) listed in
PSAs are first screened by defining the event vector, rating the MCS dependence,
setting the truncation point for the list of MCS and finally revaluation
of the results. The next step in the WPAM-II is quantification.
The organisational factors that may affect each candidate parameter group
are first identified through WPAM-I. Next, the success likelihood
index methodology(SLIM) is used to find new frequencies for each minimal
cut set. This involves the determination of importance weights through
the use of analytical hierarchy process(AHP), the determination of performance
ratings using tools such as behaviorally anchored rating scales and the
determination of calibration constants for the probabilities of similar
and dissimilar events. A point to note about the WPAM is that it
may double-count the dependence of the organisational factors, if the HEPs
used have already taken into the account the underlying factors, which
may at times be implicitly modeled.
The techniques to identify and quantify organisation error are also
documented in literature[12, 49]. Elisabeth Pate-Cornell and Robert
G. Bea (1992)[49] linked the probabilistic risk assessment inputs to decisions
and errors during the three phases of design, construction, and operation
of off-shore platforms. They also assessed the contribution of different
types of error scenarios to the overall probability of platform failure.
BACK to TOP
BACK to HOME
4 Conclusions
In this section, the various existing techniques and data sources for
probabilistic risk analysis have been discussed. Most research done
in this field has been dealing with nuclear power plant, thus in using
the data from the sources given, one should exercise discretion.
BACK to TOP
BACK to HOME
Any comments, please e-mail me at thk@pacific.net.sg