ASSESSMENT OF HUMAN ERROR IMPORTANCE IN PWR PSA · ASSESSMENT OF HUMAN ERROR IMPORTANCE IN PWR PSA KAVEH KARIMI1,*, FARAMARZ YOUSEFPOUR2, ALI ABBASPOUR TEHRANIFARD1, MOHAMMAD POURGOL-MOHAMMAD3

ASSESSMENT OF HUMAN ERROR IMPORTANCE IN PWR PSA

KAVEH KARIMI1,*, FARAMARZ YOUSEFPOUR2, ALI ABBASPOUR TEHRANIFARD1, MOHAMMAD POURGOL-MOHAMMAD3

1Department of Nuclear Engineering, Science and Research Branch, Islamic Azad University, Tehran, Iran

2Nuclear Science and Technology Research Institute, Tehran, Iran 3 Sahand University of Technology, Tabriz, Iran

*E-mail: [email protected]

Received August 23, 2013

Pilot errors have undeniable role in the total risk of engineering systems including Nuclear Power Plants (NPPs). Analysis of this role is the subject of this paper. Human actions affecting the risk of a typical pressurized water reactor are identified using Probabilistic Safety Assessment (PSA) model. And Human Error Probabilities (HEPs) for each action is estimated by using Human Reliability Analysis (HRA) techniques. Thereafter, human actions are incorporated in the PSA model. Finally the total risk of NPP is calculated which becomes 1.99 E-05 per reactor critical year in terms of Core Damage Frequency (CDF). Sensitivity analyses showed that highly reliable operator actions, defined to be actions with HEP of one order of magnitude lesser than normal value, would decrease the CDF up to 12.76% while, perfect human actions, defined as those actions with zero HEP, show a 14.29% decrease in the CDF. The interesting point found in the sensitivity analyses is that less reliable actions would increase the CDF up to 563%. Just similar to highly reliable actions, less reliable actions are those actions with HEP of one order of magnitude higher than the normal value. With no credit to all operator actions, i.e. setting all HEP=1, CDF would rise up to 3.65E-2 per reactor critical year.

Key words: Human Reliability Analysis (HRA), Probabilistic Safety Assessment (PSA), Nuclear Power Plant Risk.

1. INTRODUCTION

Probabilistic Safety/Risk Assessment (PSA/PRA) is a powerful tool for investigation of risk in Nuclear Power Plants. The history of the PSA methodology goes back to the Reactor Safety Study (RSS) (also called WASH 1400) which was conducted by Professor Rasmussen in 1974 [1]. However nuclear utility community came to a consensus about this methodology only after TMI accident occurred in 1979. Predictions of that study attracted the attention of nuclear industry by the occurrence of TMI accident [2].

Rom. Journ. Phys., Vol. 59, Nos. 7–8, P. 873–883, Bucharest, 2014

Kaveh Karimi et al. 2

874

Now again just after Fukushima accident, PSA methodology is attracting much more attention [3]. Beside the methodology proposed in the RSS, it was the role of human actions in the progression of accident in TMI that was important [4].

In this study we have tried to investigate the role of human actions in the total risk of a PWR. This is done by using PSA techniques. Human actions affecting the risk of a typical pressurized water reactor are identified using a Probabilistic Safety Assessment (PSA) model. Then by using Human Reliability Analysis (HRA) techniques the Human Error Probabilities (HEPs) for each action is estimated.

Thereafter, human actions are incorporated in the PSA model. And, the total risk of NPP is calculated in terms of Core Damage Frequency (CDF), as a metric for risk [5].

Sensitivity analyses are performed to investigate the role of human errors in the whole spectrum of risk.

2. MATERIALS & METHODS

The goal of Human Reliability Analysis is to support PSA in identifying and assessing risks associated with complex systems. PSA, in conjunction with HRA, affords analysts the ability to look at sequential as well as parallel pathways that generate risk, including the human contribution to that risk [6].

In analyzing human actions many classifications have been proposed in HRA literature, but three categories of HAs, based on their timing with respect to accident (or IE), are commonly used in PSA studies which are as follows [6], [7] and [8]:

– Category A: Pre-initiating event interactions (also called routine actions) (e.g., maintenance errors, testing errors, calibration errors);

– Category B: Initiating event-related interactions (e.g., human errors causing system trip, human errors causing loss of power); and

– Category C: Post-initiating event interactions (also called emergency actions) (e.g., actuating a manual safety system, backing up an automatic system).

In this study only categories A and C are considered in human reliability analysis because Category B actions are assumed to be already considered in the generic data used for initiating event frequencies. In the other word, they are implicit in the selection of initiating events and contribute to their total frequency.

Category A human actions, also called "Pre-Initiators", can take place before any initiating events or plant trips occur, owing to plant personnel inadvertently disabling equipment during test, maintenance or calibration activities. Often they leave the standby equipment in an unavailable situation. These errors are analysed using THERP methodology [9] with some modifications.

3 Assessment of human error importance in PWR PSA

875

Category C human actions are usually the most important ones, and have the dominant effect on plant safety. They are analysed using SPARH methodology [6] which is a good choice when insufficient information is faced up or simplification is required.

2.1. PROCEDURE FOR EVALUATION OF CATEGORY A HUMAN ACTIONS

Category A human actions are explicitly modeled and are usually included in the system fault trees. One Category A human action is modeled for each train of the related system, or other logical grouping of equipment, and account for different procedures and equipment manipulations at this model level.

For quantification purposes, THERP approach [9] is used, with considering a Basic HEP and two recovery factors.

A basic HEP (BHEP) of 0.03 was selected as a conservative HEP for type A human errors. The BHEP of 0.03 does not include any recovery factors (RF), and represents a combination of a generic HEP of 0.02 assessed for an error of omission (EOM) and a generic HEP of 0.01 assessed for an error of commission (ECOM), with the conservative assumption that an COM is always possible if an EOM does not occur. The estimated HEP that is used for PRA model considers the recovery factors and dependence effect on the BHEP.

Two Recovery Factors have been identified for every pre-accident error: – The results of maintenance and checking of the protections and interlocks

of the safety trains are entered into the maintenance logs of the safety systems at the workplaces of Shift Supervisors, the log for recording the checks of protections and interlocks of the equipment operated from the MCR (and into the repair personnel log, if repair work was performed). Hence, the first recovery factor (F1) is the check performed by the supervisor whose personnel performed the work.

– Based on the maintenance results and reports by Shift Supervisors, the Shift Technical Advisor makes a summary entry in the safety system maintenance log about the maintenance performed. An entry is made in the operating log of the Shift Supervisor, which confirms that the system is in correspondence with the reactor plant operating conditions. Hence, the second recovery factor (F2) is the check done by the Shift Supervisor.

– Middle level of dependency between Shift Supervisor and shift personnel Low level of dependency between Shift Supervisor and Shift Technical Advisor.

– The THERP data tables provided are used to assign probabilities to each operator action or subtask. Each Category A HEP is a mean value that is calculated from the median values of THERP by assuming a log-normal distribution with an EF = 5.

Kaveh Karimi et al. 4

876

2.2. PROCEDURE FOR EVALUATION OF CATEGORY C HUMAN ACTIONS

Human errors in this category are usually the most important ones, and have the dominant effect on plant safety. For each accident sequence under analysis HRA is performed in following steps: STEP 1: Task identification All human actions having an important effect on mitigating the accident are identified. Only human actions that can have significant impact on the progress of the accident sequence are considered for further qualitative and quantitative analysis. STEP 2: Qualitative analysis In this step, each of the identified human actions is clearly described. Then some tasks might need to be decomposed to several subtasks. Most of the tasks or subtasks, consist of a diagnosis phase followed by an action phase. STEP 3: Quantitative analysis For the sake of simplifications needs for this study and its abilities, SPAR-H method [6, 10] is used for quantitative analysis, which is appropriate for such cases. This method can be used for assigning error probabilities to human actions, and evaluating the level of dependencies between human actions. For each human action, the information from the accident sequence and the systems analyses are utilized to assign appropriate values to the PSFs. Then, SPAR-H method is used to calculate the corresponding human error probability for the task. Each phase of the task, i.e. the diagnosis phase and the action phase are evaluated separately, and the resultant probabilities are then combined as a single HEP. In accordance with SPAR-H procedures, the quantification of each task is performed in the following steps: • Evaluate each PSF for Diagnosis. • Evaluate each PSF for Action. • Calculate Task Failure Probability Without formal Dependence (PW/OD). • Dependency Analysis. STEP 4: Incorporation into PSA Models The identified human error events must then be incorporated into PSA logic models, i.e. Event Trees and Fault Trees, with their calculated HEPs. Although it is possible to incorporate a human error event as a top event of an event tree, most of them are expected to appear in their corresponding fault trees, as basic events.

SPAR-H method uses base error rates modified by eight PSFs to calculate the HEPs for human actions. However for our case all eight PSFs needn’t to be considered. Thus, it is decided to use only 3 of these PSFs in our quantitative analyses: Available Time, Stress & Stressors, and Complexity.

5 Assessment of human error importance in PWR PSA

877

It is assumed that all the tasks done in MCR are performed by the same team of operators. This assumption is used in evaluating the dependency level between operator tasks. Generally, manual backup for automatically actuated systems are excluded from the analysis. According to the guidelines provided by [11], it is reasonable not to take into account this kind of actions, in early stages of HRA. There might be some exceptions to this assumption, though. Some important manual backup actions are considered in HRA, due to their large impact on final results of the total risk of the plant.

The dependency analysis for Post-Initiators follows the philosophy of SPAR-H method, which yields four dependency levels: low dependence, moderate dependence, high dependence, and complete dependence.

In dependency evaluation, if the error is the 3rd error in the sequence, then the dependency level is at least Moderate, and if the error is the 4th error in the sequence, then the dependency level is at least High, and if there are more errors in the sequence, then the dependency level “Complete” is assigned.

Once the dependency levels for each case are evaluated, the probability of task failure without formal dependence “Pw/od” is modified to account for the dependency for each dependency level. The conditional human error probability “Conditional HEP” is calculated by the following equations: • For Complete Dependence, the probability of failure is 1. • For High Dependence, the probability of failure is (1+ Pw/od)/2 • For Moderate Dependence, the probability of failure is (1+6 x Pw/od)/7 • For Low Dependence, the probability of failure is (1+19 x Pw/od)/20 • For Zero Dependence, the probability of failure is Pw/od Also, in accordance with ASME/ANS RA-Sa-2009 [10], after calculating the cutsets, the dependence between HEPs appearing in the same cutest is assessed and the values may be changed as appropriate (please see Table 1).

2.3. UNCERTAINTY DISTRIBUTION

Both HEP and conditional HEP are single parameter distributions, which are Mean values. Once the mean HEP is known, the analyst may use SPAR-H to determine an approximate distribution based on a beta distribution. The beta distribution requires two parameters, α and β. Figure 1 is showing the numerical value of α as a function of mean HEP. For example, using the SPAR-H worksheet, if one determines that the HEP has a value of 0.3, the value of α (from the curve) is 0.42. The second parameter, value of β is found via the equation:

In the case where the HEP is 0.3, value of β is found to be 0.98.

Kaveh Karimi et al. 6

878

Table 1

Dependency Condition Table

Fig.1 – Alpha as a function of mean HEP.

7 Assessment of human error importance in PWR PSA

879

3. RESULTS AND CONCLUSION

All category C human error probabilities calculated in this study using Human Reliability Analysis (HRA) techniques are summarized in Table 2.

After incorporating HRA results to the PSA model mean value of CDF is calculated to be 1.99E-5 in this study, while uncertainty analysis using Monte Carlo method reveals the 5th percentile to be 7.65E-06 and the 95th percentile to be 4.91E-05 per reactor critical year. (Please see Fig. 2)

Fig. 2 – Core Damage Frequency distribution.

In this study, by considering human actions of category A & C in the model of PSA, the role of human actions is investigated using sensitivity analyses. Human Actions of Category A is quantified using THERP approach with only two recovery factors. Category C actions are quantified using SPAR-H [6] method using only 3 performance shaping factors (PSFs).

Human actions of Category C showed more importance than Category A actions in such a way that Cat C actions are on the top ten list of Fussel-Vessely (FV) importance measure [12], defined below:

IFV = QTOP(MCS including i)/ QTOP

Operator failure to perform bleed function by Safety Depressurization Valves (SDVs) appeared in the 8th place of FV importance measure amongst all parameters.

Table 3 shows the FV and Risk Increase Factors for top ten contributing human actions based on FV.

Sensitivity analyses showed that highly reliable operator actions would decrease the CDF up to 12.76%. Highly reliable actions in this study are defined as actions with HEP (Human Error Probability) of one order of magnitude lesser than normal value.

Perfect human actions show a 14.29% decrease in the CDF, while perfect actions are defined as those actions with HEP of zero.

Table 2

Category C Human Error Probabilities (HEPs) and their parameters

Kaveh Karimi et al. -870 2Table 2 (continued)

Kaveh Karimi et al. 10

882

The interesting point found in the sensitivity analyses is that less reliable actions would increase the CDF up to 563%. Just similar to highly reliable actions, less reliable actions are those actions with HEP of one order of magnitude higher than the normal value. With no credit to all operator actions, i.e. setting all HEP=1, CDF would be rise up to 3.65E-2 per reactor critical year.

Table 3

Importance of Human Errors Based on FV

These results show the significant contribution of human actions to the total risk. They are also summarised in the Table 4.

11 Assessment of human error importance in PWR PSA

883

Table 4

Summary of the results of sensitivity analyses

REFERENCES

1. N. Rassmusen and et al., Reactor Safety Study, US NRC, 1974. 2. W. Keller and M. Modarres, A Historical Overview of Probabilistic Risk Assessment Development

and Its Uses in the Nuclear Power Industry: A Tribute to the Late Professor Norman Carl Rasmussen, Reliability Engineering and System Safety, vol. 89, pp. 271–285, 2005.

3. S. Schroer and M. Modarres, "An event classification schema for evaluating site risk in a multi-unit nuclear power plant probabilistic risk assessment," Reliability Engineering and System Safety, vol. 117, pp. 40–51, 2013.

4. Backgrounder on the Three Mile Island Accident, US NRC, 2013. 5. Modarres, Kaminskiy and Krivtsov, Reliability Engineering and Risk Analysis, A Practical Guide,

CRC Press, 2010. 6. e. a. D.I. Gertman, The SPAR-H Human Reliability Analysis Method, NUREG/CR-6883, US

NRC, 2005. 7. D. Swain, Accident Sequence Evaluation Program Human Reliability Analysis, NUREG/CR-4772,

US NRC, 1987. 8. P. M. a. G. P. J. Spurgin, A Human Reliability Analysis Approach Using Measurements for

Individual Plant Examination, EPRI NP-6560L, 1989. 9. A. G. H. Swain, Handbook of Human Reliability Analysis with Emphasis on Nuclear Power Plant

Applications (THERP), NUREG/CR-1278, US NRC, 1983. 10. Standard for Level 1/Large Early Release Frequency Probabilistic Risk Assessment for Nuclear

Power Plant Applications, ASME/ANS RA-Sa, 2009. 11. Human reliability analysis in probabilistic safety assessment for nuclear power plants, IAEA

Safety Series, No. 50-P-10, 1995. 12. S. Van Der Borst, An Overview of Importance Measures, Reliability Engineering and System

Safety, vol. 72, pp. 241–245, 2001.