Int. J. Critical Infrastructures, Vol. 1, Nos. 2/3, 2005
281
Some insights on issues related to specifics of the use of probability, risk, uncertainty and logic in PRA studies Dan Serbanescu PBMR Ltd., PRA Team, Gordon Hood Ave. 1267, lake Buena Vista Bldg., P.O. Box 9396, Centurion 0046, South Africa E-mail:
[email protected] Abstract: The paper presents the main results of the insights to a gas cooled reactor from the perspective of the following notions: probability, uncertainty, entropy and risk. Some results of the ongoing comparisons between the insights obtained from three models are presented. The approaches consider the PBMR nuclear power plant (NPP) as a thermodynamic installation and as a hierarchical system with or without considering the information exchange between its various levels. The results indicate on the convergence on some conclusions and approaches adopted for this type of reactor. Keywords: uncertainty; gas cooled reactor; risk; entropy; methods. Reference to this paper should be made as follows: Serbanescu, D. (2005) ‘Some insights on issues related to specifics of the use of probability, risk, uncertainty and logic in PRA studies’, Int. J. Critical Infrastructures, Vol. 1, Nos. 2/3, pp.281–286. Biographical notes: Dan Serbanescu is a PRA Analyst for PBMR in South Africa. He received a diploma in Nuclear Engineering from Moscow Power Institute in 1979. He received his PhD in Nuclear Engineering in the Bucharest Institute for Nuclear Technologies in 1987. He has published various papers on nuclear safety and risk analysis for CANDU plants and PRA for gas reactors and a book based on his activity as Professor Associate on master degree courses in nuclear safety and risk analysis. His research is in risk analysis methods, PRA issues for gas reactors and its use for design optimisation.
1
Overview of the methods
The concepts of risk and safety were introduced in order to improve the understanding of NPP and to solve the problems encountered during all their lifetime phases, by considering the fact that NPP are complex technical systems. As a result of this situation there are difficult problems in evaluating the uncertainties of the design models and to understand the results of operation. This is why the search for methods to improve the understanding of NPP is a high priority target. The task is more important for new designs, for which no operating experience exists. In (Serbanescu, 2003) a set of three methods was presented to be used for a comparative evaluation of a PBMR plant and some results obtained up to that moment. Copyright © 2005 Inderscience Enterprises Ltd.
282
D. Serbanescu
Each of the methods supposed that, as any modelling approach, the method is associated with a certain model of the plant. The models and methods were as follows: •
thermodynamic: model A
•
risk model: model B
•
synergetic model: called in this paper also aggregate risk model: model C.
In this paper the latest results obtained for the PBMR type of NPP using the three methods will be presented and compared. The models are using the notions of probability, risk and uncertainty as key aspects of modelling complex technical systems. The reasoning itself is done for models B and C using the specific inferences, different from the common formal binary logic.
1.1 Method A used for the thermodynamic model The model and the method used are in accordance with (Serbanescu, 2003), as represented also in Figure 1. The evaluations continued last year showed that there are no qualitative differences to be implemented in the model (Serbanescu, 2003) after the evaluation of the design changes implemented lately in the PBMR design. Figure 1
The thermodynamic – exergy model (model A) of a gas cooled NPP
Some insights on issues related to specifics of the use of probability
283
1.2 Method B used for the risk model The risk model included the use of the PRA for the evaluation of the design changes performed in the last period of time. No basic changes to the model and the method, as defined in (Serbanescu, 2003), were identified.
1.3 Method C used for the aggregate risk model The model and the method as defined in (Serbanescu, 2003) were used for the evaluation of the main design issues, for the same issues as for the other methods.
2
Main results and conclusions
The results using the three methods were in agreement with the general outcomes as presented in (Serbanescu, 2003). However new aspects related to the use of these methods were identified as follows: •
The risk profile and the aggregate risk profile – as defined by the Lagrange Function (LF) (Serbanescu, 2003) – were updated in more detail for the updated design issues. They are shown in Figures 2 and 3 and they indicate the fact that both the design review done for licensing or for design optimisation purposes are complementary. The general approach was however to perform first the review using method B and then method C. The differences between insights based on each of those methods are shortly presented below.
Figure 2
Risk profile as per method B and the LF as per method C
284
D. Serbanescu
Figure 3
LF – aggregate risk profile as per method C
•
For aspects related to the calculation of LF for a complex hierarchical system (CHS) (Serbanescu, 1991) was used. There is a high similarity and connection between methods B and C. Their main concept is to try to evaluate the impact on risk based on the plant structure and considering the uncertainties. However method C is able to evaluate better the impact of modelling uncertainties, considered as information related, on the plant risk level.
•
Method A was also used as an early input before starting the evaluations with method B. It is also aimed at identifying the impact on the thermodynamic losses of entropy (exergy) as an indicator of the main passive features – in the understanding from (Passive System Reliability, 2002) – on the robustness of plant design. Method C might be, as shown in (PRA Procedures Guide, 1983) also used as a precursor for the plant sequences review, which is a step included in method B.
•
Due to the high convergence of their results, their specialisation for some areas of insights and their overall complementary, methods A–C might be used in a combined manner in order to derive conclusions on a specific CHS. A summary of the results using all the methods is illustrated in Figure 4.
•
There would be some principles of the combined use of all the methods, as for instance: •
use them in a certain sequence (for instance A–C)
•
perform iterations if some conclusions using various methods differ between them
•
allocate highest weights to the insights resulted using the strong points of each method, as defined in Table 1.
Some insights on issues related to specifics of the use of probability Figure 4
Table 1
285
Comparison of the results for methods A–C
Comparison of the results for methods A–C Degree of ‘activity’
Degree of ‘passivity’
Rank by method A
Rank by method B
Rank by method C
Reactor and its interface systems
Low
High
I
III
III
Reactor feedback
Medium
Plant control and support systems
Medium
III
IV
II
High
Low
IV
II
I
Brayton cycle – without reactor and with its support systems
Medium
Medium
V
I
II
Water cooling systems and their support systems
Medium
Medium
II
III
IV
System
286
D. Serbanescu
•
The rule of combination of the methods would rather use the approach of the probabilistic logic (Patrick Suppes probabilistic Methaphysics, 1984). It means in short that the causal inferences done for method A using binary formal logic will be replaced with probabilistic logic formulations, which will consider the possibility that a certain conclusion has a certain degree of uncertainty, as in the methods B and C.
•
Specific insights as shown in Figure 4 and Table 1 highlight the following aspects: •
It is desirable to evaluate in a combined manner using both methods A and B the impact on the plant of changes on systems with higher degree of passivity (as defined in (Passive System Reliability, 2002)). However, by using best estimates of the evaluation of the degree of passivity of dominant systems, i.e. by considering that the uncertainties in the models B and C are kept to lower value than in the initial calculations, then the results of those methods will show the fact that the decrease in the plant risk impact for a certain series of changes is not dominated by the highly passive systems. This induces the need to improve with priority some active and/or support systems to other systems, considering in the meantime the assumptions made for the modelling.
•
The uncertainty introduced by a high level informational type of system might be evaluated in the best manner by using method C. There is a generic situation in any detailed calculation, when method C indicates on the high impact of the plant control, as a hardware sand human procedural system.
•
Plant control and Brayton cycle systems, including their support systems, have also to be in the focus of any new change, assuming that the passivity of various systems is not triggered by them.
Further evaluations and more detailed extensive use of the three methods is forseen for the next series of the PRA type applications during the PBMR design phase.
References Passive System Reliability – A Challenge to Reliability Engineering and Licensing of Advanced Nuclear Power Plants, NEA/CSNI/R (2002) 10 June, Proceedings of an International Workshop, Cadarache, France, 4–6 March. Patrick Suppes Probabilistic Methaphysics (1984) Basil Blackwell Publisher Ltd., England. PRA Procedures Guide. A Guide for the Performance of Probabilistic Risk Assessments for Nuclear Power Plan (1983) USNRC, NUREG/CR-2300, February 1. Serbanescu, D. (1991) ‘A new approach to decision making in different phases of PSA studies IAEA-SM-321/35’, in Serbanescu, D. (Ed.): IAEA – Conference on Use of Probabilistic Safety Assessment for Operational Safety PSA’91, Vienna, IAEA. Serbanescu, D. (2003) ‘Risk, entropy, synergy and uncertainty in the calculations of gas cooled reactors of PBMR type, ICAP3233’, ANS International Conference on Advances in Nuclear Power Plants, ICAPP03 Cordoba, Spain, May 4–7.
