ML23279A092
| ML23279A092 | |
| Person / Time | |
|---|---|
| Issue date: | 09/25/2023 |
| From: | Chang Y, Jing Xing NRC/RES/DRA/HFRB |
| To: | |
| References | |
| Download: ML23279A092 (1) | |
Text
Use Simulator Data to Assess Uncertainty of Time-Required for Human Reliability Analysis Y. James Chang, Jing Xing U.S. Nuclear Regulatory Commission Abstract In human reliability analysis (HRA), the time-required for operators to complete a task often contributes greatly to human error probability for the time-critical operator actions. The challenge is that the time-required to complete a task is affected by many factors, such as the scenario and context of which the task is performed and the crew-to-crew differences. The Integrated Human Event Analysis for Event and Condition Assessment (IDHEAS-ECA) uses probability distributions to represent the uncertainty of the time-required. Guidance is needed for HRA analysts to develop the probabilistic distribution when data of operators task completion time are spare. This paper analyzed the nuclear power plant operators time-required data in simulator exercises to support the guidance development. The data analyzed include those from Halden simulator experiments and operator simulator data in the nuclear power plants of the United States and other countries. This analysis results provides a data-driven technical basis for developing the IDHEAS-ECA guidance on specifying the probability distribution of time-required.
1 Introduction The General methodology of Integrated Human Event Analysis System (IDHEAS-G) [1] human reliability analysis (HRA) method quantifies the human error probability (HEP) of a human failure event (HFE) as a function of two HEPs: the error probability attributed to the uncertainty of time sufficiency (Pt) and the error probability attributed to the failure of macrocognitive functions (Pc).
Pt is calculated by the convolution of the probabilistic distributions of time-required (Treqd) and time-available (Tavil). Where, Treqd is the time that the operators taken to complete the required tasks, and Tavil is the time that the system parameter(s) would exceed the safety thresholds that determine the HFE success and failure.
This paper is one of efforts to provide data and technical basis for the United Stated Nuclear Regulatory Commission (NRC) effort to develop guidance on specifying the probabilistic uncertainty distributions for Treqd and Tavil. The results of this analysis supported a recent NRCs work to recalculate the HEPs of a set of selected HFEs in NRCs Standardized Plant Analysis Risk (SPAR) models using IDHEAS-ECA method [2]. In this SPAR HEP recalculation work, it was not practical for the analysts to collect the plant-specific and scenario-specific Treqd data.
The analysis documented in this paper provided data basis for the Treqd probabilistic distributions used in the SPAR HEP recalculation work, while Tavil was treated as constant.
2 Data Analysis Literature review was performed to understand the relevant works. The literature review concluded using lognormal distribution to represent Treqd uncertainty. The justification is discussed in section 2.1. Simulator data used in this analysis was from the literature review and international collaborations. Depending on the study purposes, the data election criteria and the data analysis methods could be different. The differences are discussed in section 2.2. Section
2.3 provides the results from a preliminary analysis. Section 3 discusses considerations of using the results of Treqd uncertainty for IDHEAS method.
2.1 Use Lognormal Distribution for Uncertainty Distribution Lognormal distribution was used by EPRI Operator Reliability Experiments (ORE) [3-5], Korea Atomic Energy Research Institute (KAERI) analysis of its Human Reliability Data Extraction (HuREX) [6] databases, and the Pacific Northwest National Laboratory (PNNL) analyses of the ORE data [7] to represent Treqd distribution. This paper uses lognormal distribution for Treqd uncertainty distribution based on the above studies. The following discusses the reasons that the above studies selected lognormal distribution to represent Treqd uncertainty.
ORE [5] conducted simulation experiments with a primary objective to verify the Human Cognitive Reliability (HCR) correlation [8] and modify it as needed. A revised time-reliability correlation (HCR/ORE) was developed to replace the original ORE correlation, which categorizes human activities into skill, rule, and knowledge (SRK) based. The ORE/HCR used cue-response types to categorize human activities.
ORE concluded that a standard probability distribution could be used to represent the operator response time. HCR originally used a three-parameter Weibull distribution to represent the distribution for its flexibility, acceptance in reliability engineering applications, and having the correct overall shape with a small number of parameters. However, ORE also pointed out that there was no theoretical reason for choosing the three-parameter Weibull distribution, and any other distributions that fit the data as well as Weibull distribution would be acceptable. Because most of the ORE data fit the lognormal distribution well, and the Lognormal distribution was available in majority of existing statistical techniques, ORE used Lognormal distribution.
KAERI analyzed the operator time in a full scope simulator of a fully digitalized main control room [9]. The analysis included diagnosis time and action execution time. Operators followed the computerized operating procedures for both diagnosis and action execution. KAERI used the R statistical software to perform the analysis. The Normal, Lognormal, Gamma, and Weibull distributions were fitted to the data, and the goodness-of-fits between the fitted distributions were compared using Aikake's Information Criterion (AIC) and Bayesian Information Criterion (BIC). A low value of AIC or BIC implies a high fitness for the data. The analysis concluded that Lognormal distribution could be preferred distribution for both diagnosis time and execution time. The AIC and BIC scores are shown in Table 1.
Table 1 Goodness-of-fits of diagnosis time in fully digitalized control room [9]
Lognormal Normal Gamma Weibull Diagnosis AIC 173.6 181.5 175.0 181.0 BIC 176.8 184.7 178.1 184.2 Execution AIC 173.0 181.1 175.0 180.6 BIC 175.9 183.9 177.8 183.4 PNNL used AIC procedure to fit ORE data using four different distributions: Lognormal, Weibull, Exponential, and a truncated Normal distribution [7]. The results, as shown in Table 2, indicated that Lognormal distribution fitted the overall data better than other distributions.
Table 2 PNNLs best fit analysis results 2.2 Data and Data Validity This analysis is to inform the guidance development on specifying Treqd distribution to calculate Pt in IDHEAS-ECA. One thing to note is that IDHEAS-ECA distinguishes two HEP contributors:
Pt (HEP due to insufficient time) and Pc (HEP due to cognitive error). Because a consequence of cognitive error is delaying the task completion time, the long Treqd data points that considered caused by cognitive error were removed from this analysis to prevent double count the impact of cognitive error. ORE noted observations relating to cognitive errors in the experiments, saying generally, one or two deviations were recorded per crew per scenario. Not all deviations were significant to key human interactions defined for a given scenario nor necessarily significant to the principal accident portion of the scenario The (deviation) are called either slips or mistake, depending on the intent of the operator [5]. However, ORE report [4] did not identify the data points that were contributed by deviations. The authors of this paper had to make engineering judgments for the determination. Generally, a Treqd, which is relatively large and shown as an outlier in the corresponding dataset, was excluded from the analysis of this paper because of cognitive errors contribution. Table 3 shows the dataset that had data excluded from this analysis. Each column is a dataset. The data are sorted from the smallest to the largest. The data points in red (the largest value of each dataset) were excluded from this analysis. The values of these data points are about two or more times greater than the data points immediately in front of them. The data points in the lower end indicating the operator tool short cut to perform the actions were not excluded from the data analysis. This because the operator took a much shorter time than the other crew to complete the actions successfully. Therefore, it is considered the cognitive error is not involved in these actions. As a result, their Treqds satisfy the Pt definition. Including these data points increased the variability.
Table 3 The data points excluded from the analysis because of suspect of involving cognitive errors. Where HI (Human Interaction) datasets were from ORE, and INTL is from international HRA empirical study.
Seq/seond HI2B1-1-1 HIB2-2-1 HI1P2-2-2 HI1P2-3-1 HIP2-6-4 HIP2-7-5AHI3B1-2-1 HI3B1-2-4 HI2P1-1-1 HIP3-3-3 INTL-LOFWINTL-SGT 1
58 2
18 60 3
105 1
43 7
276 359 8
2 85 2
120 70 6
120 2
60 9
312 497 10 3
110 3
324 73 15 360 3
71 9
342 501 10 4
127 3
784 144 21 614 4
134 13 552 598 11 5
153 4
941 183 28 1311 5
185 13 564 605 13 6
204 5
1529 495 105 8
190 15 582 627 16 7
240 5
8 303 16 660 705 16 8
293 5
12 563 17 720 717 18 9
325 6
24 19 1110 988 19 10 343 8
27 1122 2120 20 11 368 23 38 1236 25 12 686 41 1428 28 13 103 2454 42 14 93
2.3 Uncertainty Analysis 2.3.1 General Data Uncertainty Range Figure 1 shows the plots of median and error factors (EFs) of the Treqd datasets. The outlier data discussed in section 2.2 were excluded from calculating the medians and EFs. The median and EF were calculated using Microsoft Excel. The medians were calculated by applying the MEDIAN function to the dataset. The EFs were calculated by first calculating the nature logarithms of each data point of a dataset. Then the mean (mu, µ) and standard deviations (sigma, ) of the nature logarithms were calculated using the AVERAGE and STDEV functions.
The mu and sigma specify the lognormal distribution. Sigma can be converted to EF, which is defined as the 95th percentile divided by the median, or the median divided by the 5th percentile of a lognormal distribution.
The data shown in Figure 1 include all data available to this analysis. Data presented at this level is for an overall visualization. It is not practical for analysis. The plot did show that a data point, which has an EF of greater than 14, is an outlier. The data point requires further analysis.
Figure 1 An overview of the medium Treqd and error factor plots of all data points.
PNNLs analysis of ORE data [7] suggested that the EF was between 1.6 and 2.4, corresponding to the sigma values of.28 and.45, respectively. PNNL suggested that, in applications, the EF range could be used a prime value and modified by context-specified factors.
The above two analyses cannot be compared because the methods to assess the uncertainty were different. In addition, lumping all data points together to assess an overall uncertainty is considered not useful. Making the data useful would require making meaningful classification of human responses. Sections 2.3.2 and 2.3.3 discuss two different classifications.
2.3.2 Cue-Response Categorization ORE study was set out to evaluate whether classifying human tasks based on SRK, as proposed by HCR, was useful for time variability predictions [10]. ORE did not find the SRK
classification was useful. Instead, ORE study identified that a cue-response classification system was useful. ORE study [4] provided quantitative information about the uncertainty distributions of three cue-response types. The three types applicable to boiling water reactors (BWR) and pressurized water reactors (PWR). However, the uncertainty distributions for these two types of reactors are different, as shown in Table 2.
Table 4 The uncertainty ranges of cue-response types identified in ORE study Where:
CP11: Response following a change in the plant state that is indicated by an alarm or value of a monitored parameter. For example, response to a spurious pressurizer spray operation in a PWR.
CP2: Response following an event that gives rise to a primary cue (as in CP1) that has to be achieved when a parameter is exceeding or can be seen not to be maintainable below a certain value. For example, initiate residual heat removal when the suppression pool (SP) temperature exceeds 95 °C in a BWR. These human interactions involved a waiting period after the primary cue in order to reach a determined plant state.
CP3: Response following an event that gives rise to a primary cue (as in CP1) that has to be achieved before some plant parameter reaches a critical value. For example, initiating Standby Liquid Colling System (SLCS) before SP temperature reaches 110 °F in a BWR. This critical value can be regarded as a secondary cue.
2.3.3 Diagnosis and Execution categorization This section discusses classifying the human tasks into diagnosis and action execution. After the implementation of the symptom-based procedures, the operators would follow the diagnosis procedure to respond to an emergency, the transfer from the diagnosis procedures to event response procedures once the causes of the emergency are identified. The diagnosis procedures and the event response procedures represent the diagnosis and action execution, types of task respectively.
2.3.3.1 KAERI APR1400 Analysis KAERI analyzed operator data of the advanced power reactor with 1400 electric megawatt (APR1400) simulator exercises. The analysis included diagnosis time and execution time (i.e.,
implement the function (event) response procedures). The tasks were performed in a fully digitalized control room, including computerized procedures.
Diagnosis Time Analysis 1 EPRI classified human actions into Types A, B, and C corresponding to pre-initiator, initiator, and post-initiator actions. The Type C actions were further classified into that the operators responses are dictated by operating procedure (Type CP), and those typically represented the recovery of failed equipment or realignment of systems (Type CR).
5th percentile Average 95th percentile 5th percentile Average 95th percentile CP1 0.4 0.7 1
1.9 3.2 5.2 CP2 0.2 0.58 0.96 1.4 2.6 4.9 CP3 0.59 0.75 0.91 2.6 3.4 4.5 CP1 0.26 0.57 0.88 1.5 2.6 4.3 CP2 0.07 0.38 0.69 1.1 1.9 3.1 CP3 NA 0.77 NA NA 3.5 NA Error Factor BWR PWR Cue-Response Type Plant Type Sigma
Thirty-six simulation records were analyzed to assess diagnosis time. In the analysis of diagnosis time, the records where failures from a misdiagnosis or simulator problems were observed were not included in this analysis. Therefore, the diagnosis time from this KAERI analysis is consistent with the Pt of IDHEAS-ECA. The mean (mu, µ) and standard deviation (sigma, ) of a lognormal distribution (in minute) from are 2.044 and 0.33, respectively. The standard deviation of is 0.039 [9]. So, the 5th and 95th percentile of are 0.266 and 0.394, respectively. The corresponding 5th percentile, mean, and 95th percentile of EF is1.5, 1.7, and 1.9, respectively.
Execution Time Analysis Thirty simulation records were analyzed to assess the time uncertainty for action execution. The
µ and for the lognormal distribution (in minute) are 2.545 and 0.318, respectively. The standard deviation of is 0.041 [9]. So, the 5th and 95th percentile of are 0.251 and 0.385, respectively. The corresponding 5th percentile, mean, and 95th percentile of EF is 1.5, 1.7, and 1.9, respectively.
Table 5 The uncertainty bound of sigma and corresponding error factor of KAERI APR1400 study [9]
2.3.3.2 International HRA Empirical Study and other sources The international HRA empirical study [11, 12] provides detailed documentation of operator performance information in the simulated scenarios. This paper obtained the raw timing data from the researchers conducting the experiments. This paper discusses the loss of feedwater (LOFW) and steam generator tube rupture (SGTR) scenarios, performed in the international HRA empirical study. Each of these two scenarios has two variations: base and complex scenarios. Each scenario was run by between 10 to 14 crews, except a LOFW scenario from a different source that was run by 17 crews. The analysis of the Treqd in these exercises is shown in Table 6. Except Items 3 and 4, all other items were from the international HRA empirical study.
Each dataset in Table 6 has clearly identified cue and response to specify the beginning and the end points to collect Treqd. For a diagnosis task, the cue (i.e., the beginning) usually is an alarm or multiple alarms occurred simultaneously. The response (i.e., end point) is exiting the diagnosis procedures to enter the event response procedures. For an action execution task, the cue is entering the event response procedures. The response (i.e., end point) is the required tasks were performed. However, not all desired data points have data. In such situation, the points with timing data and can be used to surrogate the beginning or end point were used. For example, in Table 6, Stop RCP and First manipulation in ES-0.1 are the first action that should occurred soon after entering the event response procedure. Their times are used to surrogate the exit of the diagnosis procedures to enter an event response procedure.
The international HRA empirical study ran base (simple) and complex versions of the same events. The data inform the impacts of task complexity on the Treqd. Table 6 indicates that complexity increases the median response time but did not have visible impacts on the error factors.
Table 6 The median T(reqd) and error factors in the international HRA empirical study and another data source 5th percentile Average 95th percentile 5th percentile Average 95th percentile Diagnosis 0.266 0.33 0.394 1.5 1.7 1.9 Execution 0.251 0.318 0.385 1.5 1.7 1.9 APR 1400 Plant Type Cognition Type Sigma Error Factor
Further analysis of the data in items 5 and 6 above (the two items have larger error factors than the other items). Table 6 shows their raw data. It shows that both datasets have an outlier data point (170 and 441 seconds in Items 5 and 6, respectively) on the lower end. It suggested that these two crews (two different crews) did not follow the norm in implementing the procedures.
Removing these two outliers, the EF of items 5 and 6 becomes 1.5 and 1.7, respectively.
Table 7 The changes in error factor after removing the outlier data points 3 Discussion This paper analyzed simulator data to inform the development of the IDHEAS-ECA guidance on specifying the Treqd and Tavail uncertainty. The simulator data used for this analysis is benefited from EPRI making the ORE reports available to the public, HRA empirical studies performed to evaluate HRA methods, and international collaborations in HRA data. This paper only documented the initial analyses to inform the HRA community about the results. More data analyses will be performed to ensure the guidance to be developed with strong data basis.
Sections 3.1 and 3.2 discuss two observations from the analysis 3.2.
3.1 ORE having larger error factors The EFs calculated from ORE analyses showed larger Treqd variability comparing to other analyses. There are a few possible reasons to explain the larger variability observed in ORE analysis. Two reasons are discussed here. The first reason is that the ORE was performed in the 80s. Many of the changes were just implemented to responds to the Three Mile Island accident. The symptom-based procedures are one of the implementations. As stated in the ORE report [3]:
The "symptom-based" EOPs provide the structure for crew diagnosis and action during accident sequences. While an improvement over earlier procedures, the EOPs used at all plants reportedly had deficiencies in logic and clarity that affected crew response.
ID Scenario Cue
Response
Median (sec)
Error Factor Cognitive Type 1 LOFW base scenario Reactor trip Stop RCP 605 1.6 Diagnosis 2 LOFW complex scenario Reactor trip Stop RCP 842 1.9 Diagnosis 3 LOFW Reactor trip First manipul;ltion in ES-0.1 561 1.9 Diagnosis 4 LOFW Start a critical actioReach ES-0.1 Step 18 1716 1.6 Execution 5 LOFW base scenario Stop RCP F&B established 1933 3.7 Execution 6 LOFW complex scenario Stop RCP F&B established 1937 2.5 Execution 7 SGTR base scenario Reactor trip E-3 entered 409 1.6 Diagnosis 8 SGTR complex scenario Reactor trip E-3 entered 1197 1.5 Diagnosis 9 SGTR base scenario E-3 Entered Ruptured SG isolated 886 1.4 Execution 10 SGTR complex scenario E-3 Entered Ruptured SG isolated 1541 1.4 Execution Item 5 (sec) Item6 (sec) Item 5 (sec)
Item6 (sec 1
170 441 1183 1258 2
1183 1258 1219 1800 3
1219 1800 1776 1850 4
1776 1850 1910 1908 5
1910 1908 1957 2006 6
1957 2006 2092 2007 7
2092 2007 2162 2032 8
2162 2032 2326 2893 9
2326 2893 2505 3747 10 2505 3747 Median 1933 1957 1957 2006 EF 3.7 2.5 1.5 1.7 Original Data Outliers Removed
Several examples were noted where the lack of clarity or incomplete guidance required the crews to interpret the intent of the EOPs before they could act. This lack of clarity and/or logic leads to alternative strategies such that one cannot then expect all crews to respond in a consistent manner to a given accident situation.
The EOPs used in the other simulator exercises are expected to be in a better quality and the crews had better training on implementing EOPs than of ORE study. The EOP quality and operator training could affect the Treqd uncertainty. Better procedure quality and better operator training are expected to reduce the variability.
Another reason is the equation that ORE used to represent the uncertainty distribution. ORE used the following non-response probability equation that tended to generate a wider uncertainty distribution, i.e., a larger error factor.
P(non response) = 1
+ 1, = 1, 2,,
Where n is the number of datapoints in a dataset.
3.2 Impact of Familiarity on Treqd Variability The ORE provided a rare data to evaluate familiarity impacts on Treqd variability. In the PWR-1/Series II experiments, failure of a pressurizer spray valve was introduced into three scenarios and failure of a steam generator level instrument was introduced into others to examine the impact of familiarity impacts. The ORE report [5] only provided data about the pressurizer spray valve scenarios, as shown in Table 8. Where scenarios 1, 2, and 3 were performed in the same series at the same plant. Even though the three scenarios were different, but they all required the operator to operate the pressurizer spray valve. Scenarios 1, 2, and 3 were performed sequentially in the same series. As shown in Table 8, the median response time and the variability (sigma and EF) were trending downward. A caution is that the median time shown in Table 8 were less than 20 seconds. Whether the familiarity effects can be applied to the longer execution tasks (e.g., implement event response procedures) is still to be evaluated.
Table 8 The impact of familiarity on T(reqd) variability from ORE PWR1/Series II experiments Scenario #
Mean (sec)
Median (sec)
Sigma EF 1
18.7 15.5 0.55 2.5 2
10.4 8.5 0.47 2.2 3
8.0 7.0 0.26 1.5 3.3 Impact of Scenario Complexity on Treqd Variability As discussed in section 2.3.3.2, the international HRA empirical study performed base and complex versions of the same event. The data can be used to analyze the task complexity impacts on Treqd. For clarify, the relevant scenarios shown in Table 6 are shown in Table 9. In addition, the outlier data points shown in Table 7 were removed. Data in Table 9 showed that increasing scenario complexity tends to increase the time spent on the diagnosis procedures.
However, the data showed mix impacts on the time spent on the event response procedures.
The LOFW scenarios showed insignificant difference, but the complex SGTR scenario took much longer time than the base SGTR scenario. As for impacts on variability, the changes in EF are insignificant between the base and complex scenarios.
Table 9 The International HRA empirical study data for analyzing task complexity effect on T(reqd).
3.4 Final Remarks This paper discusses observations from a preliminary analysis of simulator data to inform the guidance development of specifying the uncertainty distribution for Treqd. Even though the data are limited, and the scenario and performance detail of the ORE data are not available, the data provided evidence that were sufficient to inform guidance development. This work can be performed largely because EPRI made the ORE data available to the public and the HRA empirical studies. Only a small portion of the data came from the NRCs collaborations with international organizations. This phenom pointed to the reality that sharing simulator data between organizations remains challenging. If the HRA community can overcome the data sharing/exchange barriers, the current and future HRA methods would benefit significantly on their data basis and estimation credibility.
References:
- 1.
Xing, J., Y.J. Chang, and J. DeJesus, The General Methodology of an Integrated Human Event Analysis System (IDHEAS-G), 2021, NUREG-2198 U.S. Nuclear Regulatory Commission, ADAMS Accession No. ML21127A272.
- 2.
Xing, J., Y.J. Chang, and J. DeJesus, Integrated Human Event Analysis System for Event and Condition Assessment (IDHEAS-ECA), 2020, NUREG-2256, U.S. Nuclear Regulatory Commission, ADAMS Accession No. ML22300A117 https://www.nrc.gov/docs/ML2230/ML22300A117.pdf.
- 3.
Spurgin, A.J., P. Moieni, C.D. Gaddy, G.W. Parry, D.D. Orvis, J.P. Spurgin, V.
Joksimovich, D.P. Gaver, and G.W. Hannaman, Operator Reliability Experiments Using Power Plant Simulators Volume 1: Executive Summary, 1990, EPRI NP-6937, Vol 1, Electric Power Research Institute, https://www.epri.com/research/products/NP-6937-V1.
- 4.
Spurgin, A.J., P. Moieni, C.D. Gaddy, G.W. Parry, D.D. Orvis, J.P. Spurgin, V.
Joksimovich, D.P. Gaver, and G.W. Hannaman, Operator Reliability Experiments Using Power Plant Simulators Volume 3: Appende\ixes, 1990, EPRI NP-6937, Vol 3, Electric Power Research Institute, https://www.epri.com/research/products/NP-6937-V1.
- 5.
Spurgin, A.J., P. Moieni, C.D. Gaddy, G.W. Parry, D.D. Orvis, J.P. Spurgin, V.
Joksimovich, D.P. Gaver, and G.W. Hannaman, Operator Reliability Experiments Using Power Plant Simulators Volume 2: Technical Report, 1990, EPRI NP-6937, Vol 2, Electric Power Research Institute, https://www.epri.com/research/products/NP-6937-V1.
- 6.
Jung, W., J. Park, Y. Kim, S.Y. Choi, and S. Kim, HuREX - A framework of HRA data collection from simulators in nuclear power plants. Reliability Engineering & System Safety, 2020. 194: p.
106235.http://www.sciencedirect.com/science/article/pii/S0951832017309857.
- 7.
Jefferson, B.A., G. Coles, J. Baweja, Y.H.J. Chang, and R. Prasad. Estimating the Contributions to Human Error Probability from the Convolution of the Distribution of Time ID Scenario Cue
Response
Median (sec)
Error Factor Cognitive Type 1A LOFW base scenario Reactor trip Stop RCP 605 1.6 Diagnosis 1B LOFW complex scenario Reactor trip Stop RCP 842 1.9 Diagnosis 2A LOFW base scenario Stop RCP F&B established 1933 1.5 Execution 2B LOFW complex scenario Stop RCP F&B established 1937 1.7 Execution 3A SGTR base scenario Reactor trip E-3 entered 409 1.6 Diagnosis 3B SGTR complex scenario Reactor trip E-3 entered 1197 1.5 Diagnosis 4A SGTR base scenario E-3 Entered Ruptured SG isolated 886 1.4 Execution 4B SGTR complex scenario E-3 Entered Ruptured SG isolated 1541 1.4 Execution
Available and Time Required. in 18th International Probabilistic Safety Assessment and Analysis (PSA 2023), July 15-23, 2023. 2023. Knoxville, TN: American Nuclear Society.
- 8.
Hannaman, G.W., A.J. Spurgin, and Y. Lukic. A model for assessing human cognitive reliability in PRA studies. in IEEE conference on human factors and power plants. 1985.
Monterey, CA (USA); 23-27 Jun 1985: IEEE Service Center; Piscataway, NJ (USA).
- 9.
Kim, Y., S.Y. Choi, J. Park, W. Jung, and S.J. Lee. Estimating Diagnosis Time of Emergency Situations in Digitalized Control Rooms (presentation slides). in AHFE 2018 International Conference on Human Error, Reliability, and Performance. 2018. July 21-25, Florida.
- 10.
Parry, G.W., B.O.Y. Lydell, A.J. Spurgin, P. Moieni, and A.N. Beare, An approach to analysis of operator actions in probabilistic risk assessment, 1992, TR-100259, Electric Power Research Institute.
- 11.
Bye, A., E. Lois, V.N. Dang, G. Parry, J. Forester, S. Massaiu, R. Boring, P.Ø. Braarud, H. Broberg, J. Julius, I. Mnnist, and P. Nelson, International HRA Empirical Study -
Phase 2 Report: Results from Comparing HRA Method Predictions to Simulator Data from SGTR Scenarios, 2011, NUREG/IA-0216, Vol. 2 (ADAMS Accession No. ML11250A010), U.S. Nuclear Regulatory Commission, https://www.nrc.gov/docs/ML1125/ML11250A010.pdf.
- 12.
Dang, V.N., J. Forester, R. Boring, H. Broberg, S. Massaiu, J. Julius, I. Mnnist, H.
Liao, P. Nelson, E. Lois, and A. Bye, International HRA Empirical Study - Phase 3 Report: Results from Comparing HRA Methods Predictions to HAMMLAB Simulator Data on LOFW Scenarios, 2014, NUREG/IA-0216, Vol. 3 (ADAMS Accession No. ML14358A254), U.S. Nuclear Regulatory Commission, https://www.nrc.gov/docs/ML1435/ML14358A254.pdf.