Human Reliability Analysis (HRA) - · PDF file innovative entrepreneurial global Arshad Ahmad [email protected] Human Reliability Analysis (HRA)

innovative ● entrepreneurial ● global www.utm.my

Arshad Ahmad [email protected]

Human Reliability Analysis (HRA)


Human Reliability Analysis

§  A structured approach used to identify potential human failure events (HFEs) and to systematically estimate the probability of those errors using data, models, or expert judgment

§  HRA provides the followings: •  Identified and defined human failure events (HFEs)

•  Qualitative evaluation of factors influencing human errors and successes

•  Human error probabilities (HEPs) for each HFE

2


Objectives of Human Reliability Analysis

§  To ensure that the key human interactions are systematically identified, analyzed, and incorporated into the risk analysis in a traceable manner.

§  To quantify the probabilities of their success and failure.

§  To provide insights that may improve human performance. Examples include improvements in the man-machine interface, procedures and training, better match between task demands and human capabilities, increasing prospects for successful recovery, minimizing the impact of dependencies between human errors, and so on.

3


Phases of HRA

1.   Modeling of the potential contributors to human error. This phase typically enlists some variety of task analysis to decompose an overall sequence of events into smaller units suitable for analysis.

2.   Identification of the potential contributors to human error. At this phase, relevant performance shaping factors are selected

3.   Quantification of human errors. At this phase, a human error probability (HEP) is calculated.

4


Parallel HRA Methods Developments

5

THERP ATHENA SLIM

SHARP HCR/ORE CBDT

HEART CREAM MERMOS CAHR

US NRC Sponsored Method

EPRI Developed Method

Popular InternationalMethod

ASEP SPAR-H

SHARPI HRA Calculator

NARA CARA

Athena: A Technique for Human Error Analysis ASEP: Accident Sequence Evaluation Program CAHR: Connectionist Assessment of Human Reliability CARA: Controller Action Reliability Assessment CBDT: Cause Based Decision Tree CREAM: Cognitive Reliability Error Analysis Method HCR/ORE: Human Cognitive Reliability / Operator Reliability Experiment HEART: Human Error Assessment and Reduction Technique MERMOS: Method d’Evaluation de la Realisation des Missions Operator pout la Surete NARA: Nuclear Action Reliability Assessment SHARP: Systematic Human Action Releiability Procedure SLIM: Success Likelihood Index Methoid SPAR-H: Standardized Plant Analysis Risk -human THERP: technique for Human Error Rate Prediction


Generations of HRA

6

US Airforce weapon safety projects at Los Alamos and Sandia

1962: Human Factor Conference: Symposium of Human Error Quantification

Techniques for Human Rate Prediction (THERP)

1975: WASH-1400 Application to Nuclear Power

1983: NUREG/CR1278 THERP

Simplified IG methods (ASEP, SPAR-H, HEART)

Cognition, Context, Commission

2G Method (CREAM, ATHEANA, MERMOS)

3G Method (MicroSaint, IDAC, MIDAS)

Dynamic PRA

innovative ● entrepreneurial ● global www.utm.my innovative ● entrepreneurial ● global www.utm.my

Success Likelihood Index (SLIM)

Method


SLIM

§  Expert judgment technique using decision analysis method §  Developed by Embrey, HRA, UK. Sponsored by NCR USA §  The likelihood of error occurring in a particular situation depends on the

combined effects of a small set of PIFs §  PIFs include: level of training, stress, design of equipment, design of

procedures §  Some PIFs have more influence than others (they should be given more

weight) §  There is a lack of empirical data on Human Error Probability (HEP) and

such data are situation specific §  Where empirical data is absent, it is best to rely on the judgment of

experts who know the tasks best §  Such judgments need to be made through systematic means

8


SLIM Method

§  Uses inputs from experts to develop a relationship between a Success Likelihood Index (SLI) and the factors influencing success

SLI = f (PIF 1, PIF 2,…PIF N)

§  SLI is a function of training, task complexities, equipment design, procedures etc

9


Terms used in SLIM

§  SLI is a numerical measures of the likelihood of task successes or failures derived from expert opinion, between 0 and 1

§  PIF is the factors e.g. quality of procedures, training, design and time available, etc

§  HEP is the probability that the functional objective of a task or task step will not be attained

§  Task is a series of steps or actions which are carried out by an operator with the intention of achieving some explicitly or implicitly defined objectives.

§  Task Step is a series of observable actions which make up a task

10


SLIM Assessment

11

SLI = Ri∑ Wi 0 ≤ SLI ≤ 100

Ri = Rating of the task on PIFi

Wi = Importance weight of PIFi

Wi∑ = 1


Steps in Carrying out a SLIM Assessment

1. Setting up initial databases i.  Identify tasks with known PIF characteristics

ii.  Group tasks into database with similar PIFs iii.  Rate all tasks in a database on these PIFs

iv.  Evaluate the relative importance weights for the set of tasks in a database

v.  For each database, combine important weights with the PIF ratings to give a SLI for each task

12



2. Calibrating the Tasks i.  Assign HEPs to tasks in database

ii.  Evaluate the best fitting relationship between HEPs and SLIs for tasks in each database

iii.  If poor fit, examine PIF rating or calibration HEPs to improve fit

13



3. Using existing database i.  Classify new tasks into appropriate existing databases ii.  Rate new tasks in PIFs in appropriate databases

iii.  Generate SLIs iv.  Convert SLIs to HEPs using calibration equations in

each database. •  A relationship Log p = aSLI + b is assumed to exist between SLIs and

HEPs. P is the probability of success and a and b are constants; •  Then using known values of HEPs and SLIs for two tasks, the equation is

calibrated (a and b are computed) and used to determine the corresponding HEP based on the computed SLI

14


Performance Influencing factors

(PIF)


Human Reliability

§  How do we measure human performance ? §  How do we measure human reliability ? §  Conditions that influence human performance have been represented

via several ‘context factors’ §  These context factors are referred to by different terms according to

method •  PSF (performance shaping factors) •  PIF (performance influencing factors) •  IF (influencing factors) •  PAF (performance affecting factors) •  EPC (error producing conditions) •  CPC (common performance conditions), and so on.

§  These factors are used as causes or contributors to unsafe, human actions in event analysis

16


Performance Influencing Factor (PIF)

§  PIF is a factor that combine with basic human error tendencies to create error-likely situations. So, it can be used to determine the likelihood of error or effective human performance.

§  PIF depends on human conditions (physically / emotionally) and the work environment or conditions

§  External factors are considered to have greater impact.

§  PIFs such as quality of procedures, level of time stress, and effectiveness of training, will vary on a continuum from the best practicable to worst possible

17


Example of Internal & External PIFs

Internal PIF §  Emotional state §  Intelligence

§  Motivation/attitude

§  Perceptual abilities

§  Physical condition

§  Sex differences

§  Skill level

§  Social factors

§  Strength / endurance

§  Stress level

§  Task knowledge

§  Training/experience

18

§  External PIF §  Inadequate workspace and layout

§  Poor environmental conditions

§  Inadequate design

§  Inadequate training and job aids

§  Poor supervision


More Example of PIFs

Policy & Organizational Culture §  Safety beliefs §  Attitudes towards blame

§  Reporting and Feedback System

§  Reward Systems

§  Third Party

§  Policy for procedures and training

§  Policies for design

§  Policies for systems of work

§  Level of participation

§  Management communications and feedback

19

Job & Task Characteristics §  Systems of work §  Maintenance

§  Control room design

§  Control panel design

§  Job aids and procedures

§  Training §  Task allocation

§  Field workplace design


Example of PIFs

Process Environment Demands §  Control room environment §  Field work environment

§  Levels of demands on personnel

§  Complexity of process events

§  Perceived risk

§  Suddenness of onset of events §  Requirements for concurrent

tasks §  Work pattern

20

Work Group Issues §  Functional interfaces §  Distribution of workload and

resources §  Clarity of responsibilities

§  Communication – internal & external

§  Authority and leaderships

§  Group planning and orientation Work Group Issues

§  Competence

§  Motivation

§  Interpersonal style §  Learning style

§  Thinking style


Example of PIF Rating

21

PIF Eval. Scale (Qualitative and

Quantitative) Procedures Physical Work Environment

WORST 1

•  No written procedures, or standard way of performing tasks

•  Not integrated with training

•  High levels of noise •  Poor lighting •  High or very low temperatures

and high humidity or wind chill factors

AVERAGE 5

•  Written procedures available, but not always used

•  Standardized method for performing task

•  Moderate noise levels •  Temperature and humidity •  range

BEST 9

•  Detailed procedures and checklists available

•  Procedures developed using task analysis

•  Integrated with training

•  Noise levels at ideal levels •  Lighting design based on

analysis of task requirements •  Temperature and humidity at

ideal levels


Example - Fatigue

22

Scale Time at Work Amount of Sleep Shift Rotation

1 Work shifts of twelve hours or longer, with few breaks and are unhappy with their working environment

Less than five hours sleep and have no opportunities to reclaim lost sleep during the week

Worker frequently change shift, start earlier than previous shifts and undertake shifts of longer than 12 hours

2

3,4,5,6

7 Workers work regular shifts of eight hours or less, have regular breaks and are happy in environment

Have at least eight hours sleep per night with regular and frequent rest days

Workers do the same shift all the time and work shifts of 8 hours or less

8

9

§  Fatigue is caused by (1) Time at Work, (2) Amount of Sleep, (3) Shift Rotation


Example- Fatigue

23

Using these scales, the task expert decides that the amount of sleep that the workers achieve and the shift rotation schedules are fairly close to the best case scenarios describes in the scales

Fatigue (3 + 8 + 7) / 3 = 6

Time at Work (3)

Amount of Sleep (8)

Shift Rotation (7)


PIF Example: Operator Behavior

24

ways by which the value or state of a given PIF can beassessed directly, or as a function of other PIFs. Support-ing evidence from psychological literature, experiments,actual events, and various HRA methodologies is pro-vided.

An influence diagram, discussed in this paper is used torepresent a set of causal relations among the PIFs. Thisinfluence diagram is supplemented by a number ofmathematical relations for a more explicit representationof such relationships. These sometimes take the form ofcorrelation or stochastic relations rather than deterministiclinks.

2. IDAC performance influencing factors

When an individual encounters an abnormal event, thenatural reaction often includes physical, cognitive, andemotional responses [7]. These three types of response alsoinfluence each other. There is ample evidence suggestingthat they also affect an individual’s problem-solvingbehavior (discussed in details in Paper 4 [2]). In additionto these internal PIFs, there are external PIFs (e.g.,organizational factors) that also affect an individual’sbehavior directly and indirectly.

The PIFs discussed in this paper are those that couldplay a tangible role in altering the course of an eventthrough their effects on the operators’ responses. Thescenarios of interest are relatively dynamic and have a timewindow of up to a few hours. Thus, the PIFs requiring arelatively long time to have effects are not considered (e.g.,learning related factors). The PIFs identified in this paperare mostly ‘‘frontline’’ factors. Those factors that have anindirect influence on operators’ response are implicitlymodeled by their influences on these frontline factors. Forexample, continuously long work hours or hard-to-adjustshift schedule could cause fatigue. In the current discus-sion, fatigue is a PIF affecting the operator’s performance.The inappropriate shift schedule and long shifts are notincluded. Operator training is another example. Trainingaffects the operator’s proficiency in handling systemanomalies; thus the proficiency (i.e., knowledge and skills)but not the training is a IDAC PIF. Of course such factorscan also be added to the list explicitly in another deeperlayer of the causal model.

A key requirement in identifying factors for use in acausal model for human errors is to have a precisedefinition for each factor, and to ensure that they do notoverlap in definition and role in the overall model. This isimportant since IDAC is primarily developed to be appliedin computer simulation. Accordingly all the rules andfactors that guide and affect an operator’s behaviors mustbe explicitly represented as computer instructions. As aresult, there are fifty PIFs (divided into eleven groups) inthe IDAC model compared with ten or even fewer PIFsused in typical HRA methods designed to be usedmanually (e.g., [8,9]). The IDAC PIFs allow a more precisedefinition in state assessment and causal mechanisms, and

enable computer rules to interpret small differences incontext which could result in visible different behaviors. Insimpler models where expert judgment is often used torelate context to behavior, it is not practical to considermore than a handful of PIFs. It is relatively easy to see thatthe larger set of IDAC PIFs can be reduced throughgrouping and/or scope reduction.Developing precise and non-overlapping definitions for

all PIFs is extremely difficult given the current state of theart, the quality, form, and availability of relevantinformation, and complexities of communication acrossdiverse disciplines in which subjects are studied often forentirely different reasons and end objectives. IDAC hasmade an attempt to meet these requirements. The fiftyIDAC PIFs are classified into eleven hierarchicallystructured groups. The PIFs within each group areindependent; however, dependencies may exist betweenPIFs within different groups. Fig. 1 shows the dependen-cies of the IDAC PIF groups.As stated earlier, PIFs are grouped into internal PIFs

and external PIFs. The internal PIFs are further dividedinto three groups: Mental State, Memorized Information,and Physical Factors. Mental State covers the operator’scognitive and emotional states. It consists of five PIF sub-groups representing different facets of an operator’s stateof mind. These five PIF sub-groups are hierarchicallystructured to represent a process of cognitive andemotional responses to stimuli, from top to bottom,including Cognitive Modes and Tendencies, EmotionalArousal, Strains and Feelings, Perception and Appraisal,and Intrinsic Characteristics. Memorized Informationrefers to the system-related information that is either

ARTICLE IN PRESS

Fig. 1. Organization of PIF groups and high-level interdependencies.

Y.H.J. Chang, A. Mosleh / Reliability Engineering and System Safety 92 (2007) 1014–1040 1015


Full Set PIF Taxonomies

§  Detailed set of PIF is developed for human factor analysis •  CSNI taxonomy (Rasmussen, 1981)

•  THERP (Swain and Guttman, 1983)

•  HEART (Williams, 1988)

•  PHECA (Whalley, 1987)

•  PSF taxonomy (Bellamy, 1991) •  Influencing factors (Gerdes, 1997)

•  K-HPES (KEPRI, 1998).

25


PIF Taxonomy for Human Reliability Analysis (HRA)

26

•  Quantification of HEP: SLIM (Embrey, 1984), PLG- SLIM (Chu et al., 1994), INTENT (Gertman et al., 1992), STAHR (Phillips, Humphreys, Embrey & Selby, 1990), and HRMS (Kirwan, 1997)

•  Analysis of errors of commission: Macwan’s PIF taxonomy for errors of commission (Macwan & Mosleh, 1994), Julius’ PIF taxonomy for errors of commission (Julius, Jorgenson, Parry & Mosleh, 1995), and A THEANA (US NRC, 2000)

•  Overall context assessment and error analysis: HRMS, CREAM (Hollnagel, 1998; Hollnagel, Kaarstad & Lee, 1999), and INCORECT (Kontogiannis, 1997)

•  Database for HRA: Taylor-Adams’ PSF taxonomy for CORE-DATA (Taylor-Adams, 1995), and Rogers’ PSF taxonomy for CORE-DATA (Gibson et al., 1998).


PIF Development for HRA (Example of Accident Modeling)

27

Full Set PIF HRA PIF

H U M A N

Cognitive Factors •  Attention •  Intelligence •  Knowledge •  experiences Psychological States •  Stress •  Burden Personal States Self-esteem Self-confidence Sense of responsibility : Morale & motivation

•  Competence •  Adequacy of

training & experience

•  Stress •  Workload

•  motivation

Candidate PIF

Knowledge, experiences, adequacy of training & experience, competence, stress, burden, fear of consequence, Morale & motivation

Grouping

•  Knowledge, experiences, training, competence

•  stress, burden, fear of consequence, •  Morale & motivation

Final Selection & Structuring

•  Training & experiences •  …

Situational characteristics of accident model

Selection Criteria for HRA

J.W. Kim, W. Jung (2003). A taxonomy of performance influencing factors for human reliability analysis of emergency tasks, Journal of Loss Prevention in the Process Industries,16,479–495


PIFs for HRA (Example of Accident Modeling)

PIF Group Representative PIF Sub-items Human 1. Training & Experience Adequacy of training (frequency, recent training, fidelity of simulation

program) Experiences/practices of real operating events Learning of the past events/experiences Career of the operators

Task 2. Availability & quality of procedures 3. Simultaneous goals/tasks 4. Task type/attributes

Availability Format or type Clarity of instruction and terminology Decision-making criterion Logic structure Number of simultaneous goals/tasks Priority between goals/tasks Conflict between goals Type of man–machine interaction Dynamic/step-by-step Task criticality/consequences Degree of discrepancy with familiar tasks

28

J.W. Kim, W. Jung (2003). A taxonomy of performance influencing factors for human reliability analysis of emergency tasks, Journal of Loss Prevention in the Process Industries,16,479–495


PIFs for HRA (Example of Accident Modeling)- cont’d

29

PIF Group Representative PIF Sub-items

System 5. Availability and quality of information 6. Status and trend of critical parameters 7. Status of safety system/component 8. Time pressure

Information availability (instrumentation fail/stuck) Clearness of meaning (direct indication/interpretation required/ambiguous/unreliable information) Distinguishability of information Control display relationships Value of critical parameters Trend of critical parameters (rate of change of critical parameters Number of dynamic changing variables Degree of alarm avalanche Success/fail state of safety system/component Level of trust on the system/component Number of failed/stuck components Previous operation history and current status of safety system Available time vs. required time


PIFs for HRA (Example of Accident Modeling)- cont’d

30

PIF Group Representative PIF Sub-items

Environment 9. Working environmental features 10. Team cooperation and communication 11. Plant policy and safety culture

Task location: (MCR/local CR/local area) Accessibility Clearness in role/responsibility definition Direction, type, method, protocol Standardization in instruction/information delivery Team collaboration/cooperation Adequacy of distributed workload Plant specific prioritized (or preference for/objection to) goals/strategies Attitude toward EOP/AMP training Safety/economy tradeoff Routine violations


SLIM Example


Example: Operator Performance

§  Quality of Information available to operators from control panel §  Quality of procedures §  Time available to diagnose situation and carry out actions

§  Degree of operator training

32

PIF Assigned Weight Normalized Weight Quality of Information 100 100/200 = 0.5 Training 50 50/200 = 0.25 Time Available 30 30/200 = 0.15 Quality of procedures 20 20/200 = 0.1

200 1.00

PIF Weights


Calculation of SLI

PIF Normalized Weight

Rating Weight x Rating

Quality of Information 0.5 70 35.0 Training 0.25 20 5.0 Time Available 0.15 10 1.5 Quality of procedures 0.1 50 5.0

1.0 SLI = 46.5

33


Convert SLI to Error Probability

Task SLI HEP A 80 0.001 B 20 0.01 C 46.5 ???

34

Let say there are two known task with known SLI and HEP, then we can compute the HEP for the current task with the computed SLI

logHEP = aSLI + b

Log (0.001) = -3 = a80+b Log (0.01) = -2 = a20 + b

a = -0.01667 b = -1.6667

logHEP = −0.01667SLI −1.6667

Since for this task SLI is 46.6, Log HEP = -0.1667(46.6)-1.6667 = -2.441667 HEP = 3.6 x 10-3


Problems with SLI

§  Until now, everything looks ok except the fact that you have to assign rating, which can be difficult and subjective.

§  The more difficult part is to have a knowledge on the HEP for two of the similar event and to linearly interpolate or extrapolate from those information.

35


Software

§  SLIM-SAM : SLIM Self assessment module •  Derives the SLI

•  Checks the degrees of shared variances between ratings generated by the judges

§  SLIM-SARAH: SLIM sensitivity analysis for reliability assessment for human •  Allows additional sensitivity and cost benefit analysis

36


THERP

innovative ● entrepreneurial ● global www.utm.my 38

Technique for Human Error Rate Prediction (THERP)

§  Designed for nuclear industry but had increased popularity in other industry

§  Models people as pieces of equipment. •  You either perform or you screw up.

§  Probability trees model various tasks with success/fail outcomes for each. •  Performance-shaping factors can influence success/failure

probabilities.


THERP

§  Error of Omission – leaving out a step of the task or the whole task itself

§  Error of commission•  Errors of Selection – error in use of controls or in issuing of commands •  Errors of Sequence – required action is carried out in the wrong order •  Errors of Timing – task is executed before or after when required

•  Errors of Quantity – inadequate amount or in excess

39

THERP assumes that human errors for each task can be broken down into the following categories


Performance Shaping Factor

THERP also applies performance shaping factors (PSFs) that may influence the HEP for plant-specific actions so that the HEP can be specialised to a particular plant operations the same way as component failure rates.

THERP is often referred to as a decomposition approach in that it requires a higher degree of resolution than many other techniques in task descriptions and it puts a larger degree of emphasis on error recovery.

40


THERP Procedure

§  Decompose operator tasks into elements§  List and analyse the related human operations§  Identify human errors that can occur and the relevant human error

recovery modes §  Assign nominal HEPs to each element§  Determine of effects of PSF on each element§  Calculate effects of dependence between tasks§  Model human actions in an HRA risk model; e.g., an event tree analysis§  Quantify the total task HEP


Therp

42

!!Pe =He+ PS fkk=1n∑ Wk +C

Pe is the Probability of error He is the human error probability C is a Constant PSfK performance shaping factor Wk weight associated with PSfK N is the total number of PSFs

innovative ● entrepreneurial ● global www.utm.my 43

THERP Shortcomings

§  Focuses on procedural errors that occurred prior to an accident/incident.

§  Expert judgment is highly variable.

§  More of an art than a science.


END OF LECTURE

Documents

Human Reliability Analysis (HRA) - · PDF file innovative entrepreneurial global Arshad Ahmad [email protected] Human Reliability Analysis (HRA)