Final Report, October 1979 RIING RUEAP NGwER MIT-NE-239 QUANTITATIVE METHODS FOR ASSESSING NUCLEAR FUEL CYCLE DIVERSION RESISTANCE Investigator Carolyn D. Heising Prepared by MASSACHUSETTS INSTITUTE OF TECHNOLOGY Cambridge, Massachusetts 02139 NUCLEAR EN I!NEERIN' READING Ro M- M.I.T QUANTITATIVE METHODS FOR ASSESSING NUCLEAR FUEL CYCLE DIVERSION RESISTANCE Final Report, October 1979 MIT-NE-239 Prepared by MASSACHUSETTS INSTITUTE OF TECHNOLOGY Cambridge, Massachusetts 02139 Investigator Carolyn D. Heising Prepared for Department of Energy Nuclear Alternatives System Assessment Program Washington, D.C. MIT Principal Investigator. M. Miller Energy Laboratory ABSTRACT This paper reviews available methods for quantifying nuclear fuel cycle diversion resistance. The SAI ranking approach, and diversion path analysis (HEDL), Charm method along with the works of Selvaduray (Stanford), Silvennoinen et al (Finland), Papazoglu et al (MIT), Heising (Stanford) and Fleming et al (General Atomic) were reviewed and compared. This review revealed a surprising degree of similarity and consistency between both attribute definition and quantification approaches employed, most modelers basing their work on standard utility theory. Differences are due more to the level of sophisti- cation each analyst strived for in defining their model than to any inherent inconsistency internal to risk analysis procedures, Appli- cation of the methods to a sample problem involving a comparison of both commercial and non-commercial routes to weapons usable material show close agreement of results for those methods most firmly based on utility theory. Heuristically conceived methods render conflict- ing results that are probably not reliable. Table of Contents ABSTRACT I. Introduction: Methods for Quantifying Proliferation Resistance................................ A. B. C. II. Methods for Quantifying Proliferation Resistance........ A. B. C. D. E. F. G. H. III. Proliferation and the Nuclear Power Industry......... Risk Assessment and Quantification Methods: Previous Applications to Other Risks in the Nuclear Industry and the Logic for Extension to the Proliferation Problem............ A Comment on Quantitative vs Intuitive Decision-Making: Is a Quantitative Approach to Proliferation Worthwhile?............. 1 1 3 4 6 6 Ranking vs Gaming Approaches........................ Previous Work: SAI's 1977 Review of Methods 8 (CHARM and Diversion Path Approaches)............. 10 .. (ERDA)............. Development Method SAI's Ranking G. Selvaduray's "Heuristic" Rating/Ranking 12 Approach.......................................... .. 15 Index....... Silvennoinen's and Vira's Vulnerability Bayesian Decision Analysis: I. Papazoglu et al's 22 Multi-Attribute Decision Theory Model............. Bayesian Decision Analysis: C. Heising et al's Cost-Benefit Approach to Quantifying Proliferation Risk in the Nuclear Fuel Cycle...... .. 28 Societal Risk Approach: K. Fleming et al's 36 Markov Model...................................... Comparison of Methods/Conclusions....................... 45 REFERENCES APPENDIX A: APPENDIX B: Examining Selvaduray's Method for Assessing the Safeguardability of Various Reprocessing Technologies..................................... A-1 Support Calculations for Table IX............... . B-1 B. 1 Methods for Quantifying Proliferation Resistance I. Introduction: A. Proliferation and the Nuclear Power Industry The past several years have seen the rise of concern over the connection between the worldwide spread of nuclear power and nuclear weapons proliferation. 12 This concern was expressed by President Carter in his April 1977 decision to defer commercial reprocessing and the breeder reactor.3 Also formed at that time were the International Fuel Cycle Evaluation (INFCE) ment. and the NASAP program of the U.S. govern- At the outset, so many fuel cycle alternatives were suggested4 that the only methods available for analysis seemed to be those of a quantitative ilk. However, as it became clear that the real choices for the United States involved only a few of the suggested alternatives, the need for quantitative comparisons became less urgent, Some persons questioned the desirability of using quantitative methods citing the lack of expert concensus on even such seemingly non-controversial subjects as uranium supply as invalidating quantitative approaches. Others argued that proliferation is simply too large and unwieldy a topic to be handled in any quantitative, quasi-scientific framework. This paper attempts to address some of the misgivings that have arisen concerning quantitative analysis, reviewing attempts that have been made (to this date) to place the risk of proliferation from commercial nuclear power into an analytic framework. The viewpoint ex- pressed in this paper is that of one who has personally wrestled with the proliferation problem from a quantitative perspective. The view- point expressed is therefore admittedly biased in the favor of quantitative methods. 2 Figure 1 Comparison of Average LWR Accident Risks with Fuel Cycle Facility Risks (Consequences in 50-yr. Person-REM Radiation Exposure) Probability of Accident With Consequences Greater Than or Equal to x 10- 4 V Average LWR (Wash-1400) 10 -5 10-6 Riprocessor (a ll accidents) 10-8 Reprocessor (no final filters failure) MOX Plant Cladding Waste Repository 10-10 High-Level Waste Permanent Repository 10-11 10-12 10-13 10~1 I I 100 10 | 102503 I I es4 I 5 I I 106 107 108 x: 50-yr Person-REM The study results shown are those of WASH-14005 and SAI. 1 The MOX fabrication plant services the annual requirements of sixteen 1 GWe LWRs; the reprocessor, fifty LWRs per year. The HLW repository houses the waste generated by 280' LWRs per year. For spent fuel permanent disposal, the results for the HLW repository represent a close approximation. 3 B. Risk Assessment and Quantification Methods: to Quantifying Nuclear Power Risks Previous Applications Probably the most well-known example of a risk analysis study is that of MIT's Norman Rasmussen et al. (WASH-1400).5 Fault tree- reliability assessment techniques were applied to quantify LWR accident risks. The results were then expressed as a probability-consequence curve. The pioneering role of WASH-1400 in establishing increased and public-government acceptance for the entire discipline credibility of risk analysis cannot be underestimated, for until that time, no risk analysis had been used for so important a task - previous applications had proved far too academic. In the fiel_- of nuclear engineering, use of quantitative risk analysis has been on the increase, particularly in nuclear safety applications.6-9 Use of risk techniques has spread into the broader area of fuel cycle accident risk assessment 10'1 where comparisons between reactor and fuel cycle risks have been made (see Fig. 1). Branching out still further in application are those studies that quantify nuclear waste repository risks of release to the biosphere 12,13 and those that address the even trickier problem of nuclear theft andsabotage. 1 4 ,15 In the area of safeguard systems reliability analysis, quantitative methods are being applied to help for example, plant designers to more comprehensively deal with subnational diversion risks directly in the plant design process. 1 6 ,1 7 Regulatory agencies are also keenly interested in risk assessments to help set standards in the nuclear industry. For example, the NRC is applying these methods to over fourteen specific areas related to nuclear safety;l8 these include such topics as impact of turbine missiles, an assessment of alternate ECCS configurations, development of requirements for offsite emergency response plans, etc. 4 With all the risk assessment activity in the many areas that constitute the nuclear enterprise, it is not surprising that these techniques have also been applied to analyze the nuclear proliferation problem. The logic that underlies such application is very simple based upon one possible answer to the question: "What other approach for analysis can be better?" C. A Comment on Quantitative vs. Intuitive Decision-Making: Quantitative Approach to Proliferation Worthwhile? Is a Underlying the application of quantitative risk assessment to the proliferation problem is the tacit assumption that no other analytic approach can be as useful or so well scientifically validated. Reliance on intuitive analyses is thought to mislead more frequently than quantitative analyses; the reliability of a scientific, quantitative approach is considered to be greater than for other methods of approach. The single most useful aspect of a quantitative approach is that the problem must be carefully dissected into its parts; much that is only peripheral to the question is therefore done away with leaving the most important aspects fully exposed. It must be emphasized, however, that quantitative analyses cannot substitute for the political process wherein decisions are made by reaching political concensus on controversial issues. The role quantitative analyses can play is in helping to reach that final concensus. To ignore the potential of quantitative analyses for analyzing important policy issues and to rely solely on intuitive judgment is to ignore a tool that can be of great service in resolving differences of opinion. The proper role for quantitative analysis 5 therefore, is as a tool useful in helping to reach political concensus; it is not an end in itself. As is the case with any tool, the end product of its application depends not so much on the tool's integrity as on the abilities of the person(s) using the tool; the analysis can only be as good as the ability of the analyst permits. Thus, when critics point to the (admittedly) many "bad" applications of the past, they should not conclude, as they often mistakenly do, 19,2 is inherently at fault. that it is the tool that Rather, it should be understood that analysts' abilities vary; some are more experienced, better trained and more insightful than others. Just as some artists are more talented in the use of the tools of the artistic trade, some analysts are more adept in the use of analytic tools. Quantitative analysis, like any other discipline, is neither a completely robotic task devoid of human * aspects nor a strictly unscientific, subjective pastime. Also, while some analyses (just like some paintings, books, and journal articles) may be very carefully done, others are not so well attended to. All of these considerations should be kept in mind as we now proceed to compare the various quantitative methods available for analyzing proliferation risk. Interestingly, many persons view quantitative policy analysis as a 2 while others, mainly robotic, impersonal misapplication of science within the scientific-technical community, view the practice as too unscientific, subjective and untheoretical to be considered worthwhile. 2 2 6 II. Methods for Quantifying Proliferation Resistance A. Ranking vs. Gaming Approaches Two major approaches to analyzing the relative proliferation resistance of commercial/non-commercial nuclear technologies have been identified. These are the ranking of attributes/societal risk and war By far the most prevalent in use is the ranking gaming approaches. method wherein a set of characteristics or attributes of a nuclear system are singled out that most influence its inherent resistance or proneness to national diversion of nuclear materials (proliferation risk). Then, the attributes are assessed and compared for nuclear technologies of interest (e.g., PUREX reprocessing vs. CIVEX reprocessing). Several methods for assessing each attribute and then comparing them have been developed and are discussed in this paper. These methods include those of SAI, Charm, diversion path analysis (HEDL), Selvaduray (Stanford), Silvennoinen .at al (Finland) and Heising (Stanford). Most of these methods, as will be shown, are based on ap- plications of standard utility theory and therefore share many similarities. The work of Fleming et al (General Atomic) uses a societal risk approach based on a Markov model of states and activities associated * with proliferation. An interesting approach that has yet to be directly applied to specifically analyze the proliferation problem involves a war gaming ** approach. * ** In this approach, two adversaries, possibly a non-weapons It attempts a fault-event tree analysis where probabilities of events are estimated as known distributions. A Monte Carlo simulation is then used to estimate frequencies. However, input data is given in terms of a series of attributes that correspond well with the other methods. This approach has been outlined by Benedict and Miller in a talk given at Harvard by Dr. Miller, MIT-EL78-001, February 1978. 7 state (NWS) on the verge of acquiring a nuclear explosiv-;e capability and the United States, for example, would be simulated in terms of likely action response sequences that might evolve out of the NWS's attempt to deploy its nuclear capability. The war gaming approach has mostly been applied by military analysts to simulate military operations between two or more opposing forces using rules, data, and procedures designed to depict an actual or assumed real-life situation. It is primarily a technique used to study problems of military planning, tactics and * strategy but could be applied to investigate diplomatic, international and economic-militaristic sanctions applied against an erring NWS, There are three types of war games in common use today; the training game is the least complex and is designed to provide the participants with decision making opportunities similar to those that may be experienced in combat. The operational game deals in the current organizations, equipment and tactics. It is more complex than the training game and is used to test operational plans. The research game is still more complex and is designed to study tactical or strategic problems in a future time frame. War games can be manual, computer- assisted or be totally computerized. As can be seen, it may be quite possible to employ such techniques to determine the efficacy of various competing nonproliferation policies, in particular, those that involve limitations on nuclear power facility exports or special arrangements for their operation. However, application of this approach will require careful examination of the possible policies available. See Shubik, M. and Brewer, G., Models, Simulations, and Games R-1060-ARPA/RC, Rand Corporation, May 1972, pp, 80-81. - A Survey, 8 B. Previous Work: SAI's 1977 Review of Methods (CHARM, VISA, Diversion Risk and Societal Risk Approaches) In a work sponsored by the Electric Power Research Institute (EPRI), Albert and Straker reviewed several proliferation resistance assessment methods 23 all falling within the ranking-of-attributes category. Conclusions of that work indicated that methods for the evaluation of proliferation resistance of alternative systems were "ivery preliminary". The study made three recommendations: (1) that design criteria for proliferation resistance be developed so as to determine an "acceptable" level of proliferation risk for a nuclear technology, (2) that questions in the format of the classic debate be generated with cases both pro and con represented, i.e., "can the LWR fuel cycle with plutonium recycle be designed to be acceptable from a proliferation standpoint?", and (3) that continued use of ranking approaches be followed to identify weak points in alternative systems and to indicate where engineered proliferation resistance might prove worthwhile. Three methods were singled out and applied in the SAI work; the basis for the selection was that no others had been developed which were useful at that time. These three included: method, (2) Diversion Path analysis and (3) the "Chati" SAI's own approach de- veloped for ERDA, a multiple criteria approach. is discussed in part C below. (1) The ERDA-SAI method The Charm and Diversion Path analyses are now discussed. The Charm method was outlined in 1976 24 and views proliferation risk as the attractiveness of the fuel cycle as a target for adversary 9 actions leading to the fabrication of a nuclear explosive device. This "attractiveness" is determined by a collection of characteristics which are reduced to a single number called "Charm". This factor is defined in the form of an equation: N Ti X = E i=l A iPiS iD where X = charm factor, N = number of diversion points, Ti = duration of appearance, S = self protection factor of material, such as radioactivity level, etc., A = minimum number of locations from which material is diverted, P i= D effort to process and produce a device from stolen material, and = risk of detection and the risks due to the nature of the material. The diversion path method problems based on subnational safeguards was developed by HEDL to quantify the proliferation risk of fuel cycles.25 The approach taken is quite similar to Charm. liferation risk of a fuel cycle is derived from the equation: TPWF = MAF x DPF x RMF The pro- TPWF = total proliferation weight factor, where MAF = material attractiveness factor, DPF = distribution parameter, and RMF = removal mode factor. The MAF term is heuristically defined as: MAF = MTF x MDF x RHF MTF = material type factor, where MDF = material description factor, and RHF = radiation hazard factor. Also, DPF = rQ/Ms where Q = mass of material and M5 5 required for one explosive. = material mass The removal mode factor (RMF) is intended as a means of assigning a value to the method used to divert the special nuclear material. C. Three paths were considered: RMF Path 1 Simple Theft 0.75 Substitution of inert material 0.1 Substitution of isotopic material SAI's Ranking Method Development (ERDA) SAI developed a preliminary method under ERDA auspices for evalu- ating the proliferation resistance of alternative systems as part of NASAP. An initial attempt in this program was the development of the Charm method described earlier. Further work evolved a multiple 11 attribute method26 (defined as the ERDA methodology in Albert and Straker's report 23 ). * This method defined several attributes to be assessed to determine a multiple criteria factor single factor through six indices: as opposed to a time, resource requirements, weapons production, inherent difficulty, detectability and interruptibility. the maker The last three are combined into a single factor called failure index. A weight factor, to be determined by a is then assigned to each index to rank the fuel cycles. decision Albert and Straker further elaborated on these six indices in their report by defining nine indices: time from decision to first weapon (yrs), time from material acquisition to first weapon (yrs), cost to produce first weapon (dollars), professional personnel to produce first weapon (number of personnel), material unattractiveness, material safeguardability, difficulty, detectability and interruptability (the last five indices being dimensionless). dices, quantitative "Medium", For each of the five dimensionless in- figures-of-merit and "High" values shown in were assigned to determine "Low", the final results table. The figures of merit appear to vary from 0 to 25, but no explanation is given in the work to explain the method used to assign these figures. Also, in Albert and Straker's application of the methods, quantitative weighting factors were not determined although a short discussion was made (p. 55) concerning the impact of equal weights placed on quantitative factors and the possibility that such weights would not The ERDA methodology described here was later extended and placed into a multiattribute decision theory approach by Papazoglu et al, at MIT. Their work is described in part F of this report. 12 influence the analysis applied to their particular sample problem. However, such weighting factors might need be determined for appli* cations to problems other than the one addressed by SAI. (The work by Papazoglu et al. attempts to determine weighting factors for these attributes; see part F below.) D. G. Selvaduray's "Heuristic" Rating/Ranking Approach Dr. G. Selvaduray of Stanford University has developed a method for evaluating the inherent safeguardability of various reprocessing methods as one part of his doctoral thesis. 27 In that work, Selvaduray devised a method by which some fourteen reprocessing methods for thermal ** reactors could be compared with respect to eleven parameters, one of ** which included safeguardability of strategic nuclear materials aspects. In assessing the safeguards parameter, Selvaduray defined five sub- parameters (or attributes) the combination of which could be rated according to a heuristic method he developed independently. These five The sample problem examined by Albert and Straker concerned a comparison between four reactor/fuel cycles: (1) the LWR once-through, (2) thorium prebreeder, (3) LWR with Pu Recycle and (4) IMFBR. ** The fourteen were selected froma larger sampling space of over thirty known methods. The fourteen were selected as being most representative of processes with the best possibilities for commercialization and included processes for thorium fuel although most analyzed were designed for LWR fuel. Other parameters included: (1) technical complexity, (3) magnitude of waste problems, (4) sensitivity to changing safety regulations, (5) sensitivity to change in fuel type, (6) maintenance problems, (7) stage of development, (8) reliability, (9) risk to population, (10) economic advantage and (11) decontamination factor (attribute (2) is safeguards). 13 sub-attributes included: (1) ability to extract pure Pu from process stream without additional processing necessary after diversion occurs, (2) location of reprocessor (defined as either "on" or "off"-site from one or more reactors) which impacts on transportation considerations and the size of the reprocessor that therefore affects the material flow through the process streams, (3) decontamination factor, a measure of the radiation level inherent to the process stream from which material might be diverted, (4) labor intensity, defined as the number of persons able to gain access to process streams (measured either as "high" or "low"), and (5) the number of process streams (measured either as "several" or "few"). Using these five attributes, Selvaduray created tables that rated the 48 possible combinations of the five attributes (see Fig. 2). The rating (r) assigned to each combination was allowed to vary between 1.00 to 10.00 where each of the 48 possible combinations assumed a separate rating value (e.g., 1.00, 1.17, 1.35, 1.52, etc.). ratings (r) were then multiplied by a ranking (R) These placed on safeguard- ability as one of the eleven main parameters used in Selvaduray's analysis. The formula used to arrive at a final numerical "performance factor", E, as set up by Selvaduray is: 11 r iRi 10 1R i=l i = 1, 11 attributes j = 1, 14 processes a Figure _2 Possible Combinations of Selvaduray's Safeguard Parameter Attribute 1 Attribute 2 Attribute 3 Pure Pu Extractable Location Decontamination Factor I Attribute 4 Labor Intensity j Sub-Attributes Attribute 5 Effluent Streams High Yes H High On-Site Several I No I I Of f-Site Low 15 where the denominator represents a normalization factor as derived * by Selvaduray. E. Silvennoinen's and Vira's Vulnerability Index The work of Silvennoinen and Vira in Finland 2829 used a multi- attribute approach to assess proliferation risk for several fuel cycle facilities (spent fuel storage, enrichment plants and an independent pathway followed by the non-weapons state clandestinely). The authors defined six criteria to which they assigned a quantitative value. These six criteria had been identified in a Booz, Allen and Hamilton report of 197730 and are defined as follows: (1) the minimum cost of the weapons construction once the fissile material is available, (2) the minimum time required to produce a weapon, (3) the marginal cost incurred when a commercial civil power program is amended to contribute to material for weapons production, (4) quality of the separated weapons production, (5) detectability of the conceiv- able clandestine weapons production and (6) accessibility and accountability of source material or weapons-grade material. criteria was assigned a value from 1 to 9 using a Each of these scaling * The method heuristically derived by Selvaduray can be shown to be a variation of multi-attribute decision theory where the value functions and weighting factors used by Selvaduray are linearly defined (see part III of this report). Selvaduray, trained as a chemical engineer, was not exposed to formal methods of operations research yet was able to derive what he believed to be a "heuristic" ranking method that is surprisingly well-related to decision theory methods. This adds credence to the ideas of those who claim that decision theory models well the logical thought process of the human mind. 16 method for priorities in hierarchical structures as outlined by Saaty.31 The diversion resistance of the paths was then judged in view of each criterion following the approach used by Heising. 3 2 The first step in the quantitative evaluation comprises the relative weighting of the six criteria. The criteria are then applied to each proliferation pathway in terms of the type of material that could be diverted (seven in all): (1) enriched uranium fuel in a system with no enrichment facility of its own, (2) fresh MOX fuel, (3) recently discharged spent fuel, (4) long-stored (15 yr old) spent fuel, (5) spent fuel in a final repository (>15 yr old), (6) spent fuel in a sealed final repository not intentionally retrievable, and (7) separated reprocessed plutonium (PUREX based reprocessor - no alternatives to PUREX were considered). Pathways 1, 3, and 4 were further split into two routes - military and civil facilities to obtain weapons-usable material. Each pathway was evaluated separately for each of the six criterion x as a function of the amount y of the source material. The relative values x1 -k3 were derived from data used and collected by Heising.32 The values for ,x4 -x 6 were deduced by judgmental techniques '31 based on Saaty's method. The overall vulnerability index for each of the seven materials examined was computed by summing up the ratings for each attribute x., j = 1-6 and simultaneously weighting each attribute as follows: n u(y ) =- [r j=. 1+X w.u Ji (y ) i - 1]. J 17 where = vulnerability index of material flow level y for material type j, j = 1-7 u(yi) = weighting factor on jth attribute, w j = 1 to 6 A = free parameter to normalize w. This relation was derived explicitly from Keeny's work on utility 33 theory such that n ui(yi) = 1 for all j; For the case where Z w [ = 1, A = (+ l ) - 1] = 1. 0 and the equation above simplifies j to the familiar form: n u(y) = Z W j=l u Silvennoinen et al. observed the condition y) n Z w = 1 for their sub- j=li jectively assigned weighting factors w . In their work, the authors consider the military-civilian/commercial options as a choice a NWS could make when deciding between the seven material types available to them. In their work, they assume equal preference for military vs civil in computing their final results. In a slightly different approach to calculating the indices vulnerability (or total value functions) for each of the seven material types, the authors suggest the use of "fuzzy integration" as reported in Sugeno's work.34 Instead of a weighted average, the index is taken as a fuzzy integral uF(Yi) over all the criteria x, j = 1-6: Table I Comparison of Vulnerability Indices (Total Value Functions) on Seven Material Types for Utility vs Fuzzy Integration Theory Vulnerability Index (u(yi)) Material Flow Fuel Type 1. Enriched U (MTSWU) 2. MOX Fuel (MTHM) Utility Theory * Fuzzy Integration 140 .30 .23 4600 .54 .60 30 .51 .48 1000 .54 .48 oo 3. Short Cooled Spent Fuel 30 .12 .19 (MTHM) 1000 .36 .24 4. 15 yr Old Spent Fuel 30 1000 .17 .36 .23 .28 5. Spent Fuel in Open Repository 30 1000 .14 .24 .20 .22 30 1000 .08 .12 .16 .16 6. Spent Fuel in Closed Repository 6 E j=1l w = 1. 19 f uF(Yi) a(YiX) g() X X = {x }, a(yi,X) where = ui(yi) and g(-) is a measure of integration. One interpretation of the fuzzy integral above is tantamount to maxmin algebra:34 n F V a(Yi'xj) A g(F1 ) J=1 V and A denote maximum and minimum operations, respectively. where The set F sets. i and the structure of g(-) are taken from the theory of fuzzy The measure g(F) is defined by a parameter A as follows: g (F) X Setting F = (1 + X w )-1 r = -1[ eF X and determining X exactly as was done in utility theory defines g(F) uniquely. The results of using this method were compared with the more familiar utility theory by the authors (Table I; 0 is most resistant; 1 is least resistant). Once having determined the vulnerability indices for each of the fuel types, an overall proliferation risk index (PRI) was determined for given fuel cycles that might include the seven different material types: 20 T T (1+r) max {u(yi(t)} dt + DRI 0 PRI = T (1 + r) -t where dt DRI = risk of irretrievable phase fT (1 + rD) t uy 6 (Tl)) dt where (1 + rD)-t is a time preference factor, rD is a discount rate and y 6 (T1 ) is the total amount of spent fuel disposed; T is a given time horizon, u(Y ) is the vulnerability index calculated earlier, and r is a discount rate not necessarily equal to rD. This formulation was applied to assess the difference between three LWR fuel cycle options with results shown in Table II. Table II. Overall Proliferation Risk Index (PRI) for Three LWR Fuel Cycle Options Measured* as Function of Time and Discount Rate (r) r = .04 r = .02 1. Once-Thru .41 .6 2. U-Recycle .28 .28 3. U- and Pu-Recycle .45 .45 * 0 signifies most resistant, 1 least resistant. The time horizon looked at is 20 years (T = 20); timing considerations must be taken into account because of the dependence of u(y ) on the 21 material flows which fluctuate in time. In spent fuel disposal, the discount rate (r) influences the risk index (PRI). Compared to complete recycle the once-thru alternative entails a higher proliferation risk for r = 0.02 whereas the risk is the same order of magnitude for r = 0.04. The uranium recycle case is lowest because it is assumed reprocessing is done outside the NWS so that Pu is never made available either in the stored spent fuel or in fresh MOX fuel . It is also assumed that the spent fuel is sent to the reprocessor in a short period of time while for once-thru, the spent fuel is allowed to be stored within the NWS. The authors point out that the most difficult aspect of applying either utility or fuzzy integration methods as outlined in their work is the subjective assignment of the weights (w.). They suggest a workshop approach where a number of experts would be consulted to obtain weights. They conclude that reprocessing and plutonium utiliza- tion should be retained in a more resistant system or country and that no MOX fuel be sent to a NWS. They also advocate sending spent fuel back to repositories in weapons states. In Silvennoinen's work,~ it is found that when economics are also- con- sidered along with proliferation risk in a multigoal optimization approach, an option called "M' becomes optimum: "M" is a strategy wherein most of the NWS spent fuel would be reprocessed at a reprocessor located in a weapons state, all uranium would be recycled and some MOX fuel sent back for use in the NWS. a result in the NWS. Some spent fuel would be stored as The MOX fuel used would reside in the core over 22 a longer time interval and would be shipped at irregular time periods. They show that consideration of multiple objectives can pronouncedly * impact on the optimum fuel cycle option. F. Bayesian Decision Analysis: Decision Theory Model I. Papazoglu et al.'s Multi-Attribute On the prompting of the DOE and through funding from the NASAP program, I. Papazoglu at MIT, with the advice and aid of M. Miller, N. Rasmussen, H. Raiffa and E. Gyftopoulos,35 developed a formal method for assessing the relative diversion resistance between nuclear fuel cycles. Borrowing from and extending upon the earlier work done at SAI (see Section C of this report), Papazoglu defined five attributes pertaining to diversion resistance into three sub-attributes of which the third was sub-divided resulting in seven total attributes. attributes were defined as follows: (1) the development time These or the time it takes from start to finish to develop a nuclear explosive using diverted nuclear material, (2) the warning period defined as the per- centage of the development task left to complete at the time of detection by outside agents, (3) the inherent difficulty of utilizing the technology as a source of nuclear fissile material defined further by a breakdown into three sub-attributes - the radioactivity level of the diverted material in the process, the status of scientific and technical information known about the process by the potential proliferator, and the level of criticality problem associated with the process, (4) the * Their applications have been restricted to analyses of LWR options only. 23 weapons material quality defined as the type of nuclear material di- verted (i.e., either weapons or reactor-grade plutonium, or enriched uranium (U-233 or U-235)), and lastly, (5) the development cost of the explosive construction attempt. The above attributes were developed on the basis of determining a set that would be: (1) complete, covering all aspects of concern to the problem at hand, (2) operational, be meaningful to the decision maker so that he can understand the implications of the alternatives, (3) non-redundant, avoid double counting of characteristics, and (4) minimum in number, the number of attributes should be kept as small as possible. Papazoglu et al. concluded that the major objective of the proliferator is to find a pathway out of all available that least resistance with the allows him to achieve a nuclear weapons capability. Since the desired level of capability might vary between proliferators, the method defines "aspiration levels" a given proliferator might have that might influence his choice of a suitable pathway. The major ob- jective was then defined as being equal to satisfying two sub-objectives: (1) increase tiveness the likelihood of success of the pathway. and (2) increase the attrac- Attractiveness was further divided into de- creasing the weapons development time and the monetary cost involved while the likelihood of success was divided into likelihoods of internal failure and external detection. The likelihood of internal failure was further divided into inherent difficulty of fissile material procurement and weapon design/fabrication (Fig. 2). To derive a quantitative indicator of the relative diversion resistance of a given fuel cycle, a value function V(x) was defined so Figure 2 Decomposition of Major Objective into Sub-Objectives3 5 Decrease Pathway Proliferation Resistance I Increase Increase Success Likelihood Attractiveness of Pathway .9 Decrease Development Time __T_ .W Decrease Financial Cost I Decrease Internal Failure Likelihood Decrease Inherent Difficulty in Material Procurement Weapon Development Time (x1) Monetary Cost (x5) I - Index of Inherent Difficulty (x3) Decrease External Detection Likelihood - I 4:- Decrease Inherent Difficulty in Weapon Design & Fabrication Weapons Material Quality (x4) Warning Period (x2) 25 that a dimensionless numerical indicator be calculated. varying from -1 to 0 could The numerical indicators for each attribute are then multiplied by weighting factors (X ) and summed over the total number of attributes to arrive at single numerical indicator for each a fuel cycle. Basically, the purpose of the value function is to provide a numerical measure of the relative attractiveness of the various proliferation pathways available to the would-be proliferator. Assuming preferential independence between attributes (i.e., the proliferator's value placed on each attribute is independent of the value placed on any other attribute), the value function is from utility theory : 5 V(x -x i V i(x) .E 5 i=l Because the third attribute (inherent difficulty) is divided into three sub-attributes, the above expression becomes: X1 V1 x 1) + X2 V2 (x2) V(x1-x5 3 + E X3j 3j(x3j + X4 V4 (x4 ) j=1 + X5 V 5 (x5 ) where the value functions for each of the five attributes are: 26 development time (x1 in years): V (x = - 1, 8 e~ = .49 non-crisis, .83 crisis warning period (in %): V2 (x2) = eYX 2 - 1, y = 6.93 radioactivity level (in R/hr at 1 m 1! 21 101 V31 (x31 ) 10 1-.02 _-.16 x3 1 from source): 3 4 -. 5 -. 84 1 5 1 6 -. 96 -1 status of information (x3 2 referring to levels denoted as A, B, C...I): criticality problems (x 3 3 either "High", V33 ("Low") = 0, V3 3 "Medium", or "Low"): 'Med") = -. 5, V3 3 ("High") = -l weapons material quality (x 4 refers to type of material): V4 (r.g.Pu) = -1, V4 (w.g.Pu) = -.5, V4 (H.E. U-233) = -.25 and V4 (H.E. U-235) = 0 27 and finally, for monetary development cost (106 1975 V = - 5 ax5 ; a = 2.6 x 10~4, = 1.27 non-crisis 5-5 a = 10-5, =1.9 crisis Weighting functions (Xi) derived from interviews with selected experts utilizing a standard Delphi technique were assessed for two hypothesized divertor decision environments - non-crisis and crisis (Table II). Table II. Pareto Weights on Diversion Resistance Attributes Weighting Function Attribute Non-Crisis Environment Crisis Environment Development Time A .13 .31 Warning Period X2 .15 .07 Status of Information X3 1 .38 .37 Radiation Level X32 .16 .16 Criticality Problems X33 .04 .04 .03 .01 Weapons Material Quality Development Cost 5.11 .04 This was done because it was assumed that a proliferator's choice of pathway would be influenced by the environment in which the decision takes place. In a crisis situation development cost, for example, be- comes less important to the hypothetical decision-maker while development time assumes a more important stature. In the business-as-usual 28 or non-crisis environment, the weights assume a different set of values. By allowing for possible large differences in the decision- making environment, Papazoglu et al.'s approach can account for a larger number of possible diversion scenarios. Combining the weights with the value functions allows one to calculate dimensionless numerical indicators that vary between -1 to 0 (where -1 is most resistant, 0 least resistant) for each examined fuel cycle. The importance of these final numerical indicators are as rela- tive values rather than absolute values of proliferation resistance. In summary, Papazoglu et al. have followed closely the theory of multiattribute decision theory in their methodological development. They have also gone the extra step to determine through Delphic interviews the relative weighting factors (X ) and value functions selected experts place on each attribute (xi). V(xi) This final step complished through careful construction of questionnaires that ac- represents a definite advance over other methods discussed in this report (the importance of this distinction is described more fully in part III of this report). G. Bayesian Decision Analysis: C. Heising et al.'s Cost-Benefit Approach to Quantifying Proliferation Risk in the Nuclear Fuel Cycle C. Heising et al.'s EPRI-sponsored work at Stanford University involved as cost-risk-benefit approach to assessing proliferation risk from a given fuel cycle.32 Our work basically followed a Bayesian decision analysis approach where uncertainties on both economic 29 parameters and national security were explicitly defined in terms of probability distributions. With regard to proliferation, the overall risk represented by the commercial nuclear fuel cycle inclusive of the impact that plutonium recycle and breeder introduction in the United States might have on non-weapons states (NWSs) was compared with the risk already existing from non-commercial routes to nuclear material attainment. As is shown in Fig. 3, the analysis of diversion resistance was limited to an examination of the relationship between the number of commercial reprocessors built and operating in a NWS and the likelihood of a NWS successfully constructing a weapon. The analysis does not include an examination of success of internationally applied sanctions or timing considerations that impact on sanctions success. As in the other methods discussed in this report, the emphasis lay in establishing quantitative rankings representing the relative attractiveness of available routes. This was accomplished by defining 10 attributes considered to influence the NWS in making its decision (Fig. 4). (1) These attributes included: Domestic Availability of Technology (compares the present global distribution of various technical routes; e.g., research reactors are owned and operated by over 50 countries while no commercial reprocessors are operating anywhere in the developing world); (2) Import Availability (compares the relative importability of one option over another; e.g., research reactors are far easier to import than are, say, enrichment plants); (3) Capital, Operating, and Maintenance Costs; (4) Suitability for Clandestine Operation; Figure 3 Influence Diagram Indicating Relationship Between U.S. Reprocessing Decision and NWS Technical Success in WeaponConstruction. 0 Note that the ranking method is trying to determine the relationship between the number of commercial reprocessors in a NWS and the potential for technical success of the NWS in weapon construction. The other relationships have not been included in ranking methods, mainly because analysts are technical people by training and are most expertly concerned with effects on technical NWS success likelihoods. 31 (5) Difficulty of Technical Implementation, (6) Number of Weapons Attainable from the Material Flow, (7) Quality of Weapons Material, (8) Number of Technical Personnel Required, (9) Level of Support Technology and Industry Required, and (10) Time Required to Construct a Facility. The procedure followed was to make numerical assignments on eacn key attribute. The first two attributes (domestic and import avail- ability) were treated separately because of dependence on the time period examined. The other seven attributes were compared simultan- eously to determine an overall ranking for each route deemed the A 32 "technical attractiveness" factor (Rj - see Table 7.5, p. 7-1432 Data on each attribute was extracted from the literature but was qualitative in nature (e.g., information was often expressed in terms of "high", "medium" or "low"). Therefore, if the data revealed a "high" cost for a particular route in comparison to the other alternatives, a rating of .3 = (1-.7) on a scale of 0 to 1 was assigned to indicate the relative economic attractiveness of that particular route. Thus, even qualitative statements made regarding, for example, the degree of organization or detectability of the operation were placed into quantitative terms through this assessment procedure (Table III). Table III. Qualitative Rating Numerical Assignments Associated with Qualitative Rating Numerical Assignment Very High High .9 .7 Medium Low Very Low .5 .3 .1 Figure 4. NWS DECISION TO ACQUIRE WEAPONS MATERIAL DECISION TO ACQUIRE NUCLEAR WEAPONS MATERIAL TYPE AND NUMBER OF WEAPONS DESIRED RELATIVE A TTRACTI VENESS OF ROUTES TO WEAPONS MATERIAL A TTA INMENT DOMESTIC A VAILABILITY OF TECHNOLOGY IMPORT AVAILABILITY 1 Research Reactor + MPRP Small (A) Yes Yes The decision faced by the Nth Country after it has decided to pursue the development of'a nuclear weapons capability is to determine which route (or routes) available are to be used to attain weapons-usable material as a function of the scale of capability it desires. Yes 33 Then, for each technological route to weapons usable material, the numerical assignments for each attribute were summed to arrive at an overall numerical assignment for the particular technology. This procedure corresponded to the following equation: V' N Tj.X m Z n iZ i V (x )). i i j V . j=1 where V = total numerical value calculated for route V = sum of total numerical values for each route i, J; m Z j=l NTi j = 1 to m; = normalized total numerical value for route j, m E N = 1; j=1 The xi = weight placed on attribute i; in this case xi = 1/n (equal weights assumed for each attribute); V (xi) = numerical assignment made on attribute i assuming value x (high, med, low, etc.). value functions Vi(xi) used were based on a linear hypothesis; it was assumed that the value placed on a "high" outcome for an attribute was equal to 1 minus the value placed on a "low" outcome - symmetry of values was assumed. Also, equal weights (Xi) were assumed for each * The effect of using non-linear value functions, as was done by Papazoglu et al. is discussed in part III. Table V COMPARISON OF MULTI-ATTRIBUTES OF AVAILABLE ROUTES TO WEAPONS MATERIAL ROUTE ATTRIBUTE 1 3 Capital/O&M costs (106 $) 4 Suitability for clandestine operation 5 Difficulty of technical 8 Number of technical 5 6 7 8 9 High - 100 Med. High Med. - 149 - Low Low - V. High 500-1000 V. High 1300+ V. High 650-1500 V. High 1300+ - - 200-500 3000 51 Med. 96 High High V. High V. High - Med. Low Low--Med. Med.- High Low Low Low - Low Low- Med. Med. - V. Low - Low Low Low Low V. High High V. High High V. High High V. V. High Med. V. V. High High High High High High - - V. High V. High High Low - Low - - ... - Low >1 1-4 - 10-20 - 10-20 - - 30 - - 2-20 - - 2-20 - - - 2-20 - Sev. kgs. - 1 - - 3 25 - <<1 - V. High V. High - - - Low Low Med. Med. (Med.) V. Low Med. - High V. High Low Low V. High V. High - V. High - 15 V. Low - High Med. High Med. High Med. - Med. High High - - Med. High 400 Med. personnel - - - 50 100-200 2000-10,000 2000-10,000 200-500 200-500 - -- Low Low Med. Med. Med.-High High V. V. High Med. - - Med. V. V. High Med. -Med. - - - - Low Low High V. High High High High Low - 1-3 2-3 1-3 3-4 2-3 5-8 4-6 3-5-- 3 facility +2 +2 6-8 - - -- (years) <1 5 7-11 7 16 10 Time required to construct - - 2-20 required 9 Level of support technology and industry required Low - able per year from material flow <1 material 4 10 60-120 50+ 23 Low 7 Quality of weapons 3 56 15-40 50+ 10 implementaion 6 Number of weapons attain- 2 (Data from four references: -6-8 7-11 lamarsh Starr-Zehroski SAI and Westinnho:u;I - - - High-Med. Med.-Med. Med. 23 - - 600*M- AN-- 35 attribute; each attribute was equally as important as assumed any other - no single attribute was considered more significant than the others. This was based on the assumption that, for a particular Nth country, all attributes would be considered equally as important in influencing their final decision regardless of the decision-making Results of a sample analysis environment they might find themselves in. using this method are given in Table IV based on data shown in Table V. Table IV Results of Evaluation of Alternative Near-Term Routes to SNM ** Route Attribute* 1 2 3 4 3 .9 .9 .1 4 .7 .9 5 .7 6 5 6 .1 .. 1 .5 .5 .1 .3 .7 .1 .1 .3 .7 .7 7 .7 .9 8 .9 9 7 8 9 .3 .5 .3 .5 .3 .1 .3 .1 .5 .1 .1 .3 .9 .3 .5 .5 .3 .5 .3 .3 .5 .5 .5 .5 .5 .5 .1 .1 .5 .5 .5 .5 .5 .9 .5 .3 .1 .1 .5 .5 .5 .5 10 .9 .7 .3 .3 .3 .1 .5 .1 .5 Total 6.0 Normalized .192 5.8 .186 2.4 .077 2.0 .064 2.2 .071 3.6 .115 3.2 .102 2.6 .083 3.4 .110 Attributes by Number are: (3) Capital, Operating and Maintenance Costs, (4) Suitability for Clandestine operation, (5) Difficulty of technical implementation, (6) Number of Weapons Attainable per year, (7) Quality of Weapons Material, (8) Number of Technical Personnel Required, (9) Level of Support Technology and Industry Required, and (10) Time Required to Construct Weapon. ** Routes available are: (1) Res. Reac. + MPRP, (2) Prod. Reac. + MPRP, (3) Power Reac. + MPRP, (4) Power Reac. + Com. Rep., (5) Diff. Cascade, (6) Centrifuge, (7) Aerodynamic Jet, (8) Electromagnetic Sep., and (9) Accelerator. 36 H. Societal Risk Approach: K. Fleming et. al.'s Markov Model3 6 K. Fleming et. al's ACDA - sponsored work at General Atomic used a societal risk approach to assessing the relative proliferation risk of nuclear systems. The assessment approach followed a logic similar to that applied to quantify the ldvel of safety in nuclear power plants with respect to accidental releases of radioactivity. Fleming et. al.'s pre- vious work on quantifying HTGR-related accident risk using reliability/fault tree methods provided the background for their treatment of proliferation risk assessment. The GA team defined proliferation risk as a mathemetical expression based on the following relation for risk: where R c H risk, the expected value of the consequences per unit time, = consequence of undesirable event i, = = by carrying out scenario frequency of attempt to produce c j, and p = by attempting to carry probability of successfully producing c, out scenario j. With regard to nuclear weapons proliferation, measure the undesirable consequences (c ) in the GA team decided to terms of the number of- weapons obtained in a successful proliferation attempt. They considered several scenarios' since many pathways exist involving the nuclear fuel cycle that can lead to such a successful proliferation attempt. Thus, for scenarios involving the nuclear fuel cycle, a simplified definition of risk based on the general formulation above was developed: 37 R = where H iHp = c frequency of attempt to acquire nuclear weapons by carrying out s.cenario i, probability of successful completion of P= scenario i, and c = number of weapons obtained by successfully carrying out scenario i. These parameters were further disaggregated as follows: q, c, =m f where n frequency of attempt to carry out any proliferation scenario, q= m probability that scenario i = is selected, amount of material at the specific point in the fuel cycle selected for diversion (kg), and f = number of weapons obtained per unit of material located at the diversion point (weapons/kg). These definitions lead to the final expression for proliferation risk used by the GA team: R = 11 q p miff The GA team went on to apply this expression in quantifying the level of relative risk inherent to various fuel cycles. They did not attempt to quantify the factor H (frequency of attmept to proliferate) as this is a Figure 5 Methodology for Assessment of Proliferation Resistance COMPARE RISK QUANTIFY SCENARIO PROBABILITY a WITH ALTERNATIVE FUEL CYCLES 00 39 factor independent of the fuel cycle. Therefore, their final results are expressed as conditional probabilities dependent on this factor. Further simplification of the method included the assumption that q.=1 for i = k where k is that pathway found for each individual fuel cycle that is most likely to lead to a successful completion of one or more weapons. plement this simplification, the GA team To im- quantified the entire spectrum of scenarios possible for each fuel cycle. Also, before the above equation can be applied, all diversion points needed to be identified. This was the first step in the analysis and was carried out by assuming that diversion points consist of each facility in L . fuel cycle and each transportation link where nuclear material might be found. The second step was to identify proliferation scenarios whose probabilities were then assessed in step 3 while the number of weapons obtainable from each scenario calculated separately in step 4. was This done, the relation above was then applied to calculate the relative risks (R) of each fuel cycle (Figure 5). Basically, the GA approach uses a three factor formulation of proliferation fisk (pi,fi,m.) where each of the three factors are calculated on the basis of separate models and/or assumptions. The GA team spent considerable time and effort developing an approach to assess p i, the probability of successful completion of scenario i, while basing the values for f and mi on assumptions related to a given size of the nuclear system assumed to be operating inside a NWS. a Markov transition state model where p The model for p involved can be calculated using Monte Carlo type methods based on five time factors (TpTw' dl' d 2, and ponds to six possible states of the system. (Figure 6). T s) corres- The time-dependent probability of occupying each of the six states, where state 3 is the state 40 Figure 6 STATE Markov Model of States and Activities Associated with Proliferation 1 STATE 3 STATE 2 TW PRLIf LHiATION STATk 7 AVERAGETIMETO COMP.ETE INDICATED ACTIVITY ACl1VITIES ASSUCIArED WITI TRANSI'ER FROM STATE 1U STATE T, 41 at which nuclear weapons capability is obtained, was calculated using a procedure originally developed for nuclear safety calculatior based on These methods are applicable here to the fault trees and logic models. assessment of p bacause there are limited or no statistical For the GA study, to estimate these probabilities. was based on the Markov model shown in Figure 6. data in which the logic model used Thus, to assess pi, the user of the method need only specify the average values of the completion times for each state (Table VI) which are then input to a code that cal- culates each transition state probability. To calculate f x mi = c, on a series of assumptions, approach used. the GA team chose to base their estimate none of which are inherent to the mathematical However, two of these assumptions substantially affect the results produced by the GA team; these assumptions are that: (1) of technology in the NWS is the level sufficiently high so that any difficulties in material conversion associated with radiation/criticality level or number of technical personnel required can be ignored, technology of the NNWS is cycle facilities the level of nuclear adequate for operating and maintaining all fuel assumed located in are conducted on a first (i.e.; its boundaries), and (2) calculations year of operation basis for a system assumed to consist of lOGW(e) of electrical generating capacity (i.e., time efforts are neglected). These assumptions are different from those used in the other assessment methodologies which approach the problem from the point of view of a country not yet assumed to have embarked on a nuclear power path. Therefore, it is assumed that radiation level, criticality and handling problems, etc. do affect a NWS decision between technological routes to Table VI SPECIFIC ACTIVITIES ASSOCIATED WITH NATIONAL PROLIFERATION Symbol for Average Completion Time Impact on Risk of Proliferation(a) Preparation T Unfavorable Design weapon Construct reprocessing facility Test non-nuclear components of nuclear weapon Weapons T Unfavorable Recover plutonium or HEU from fuel Fabricate nuclear components of weapon Assemble weapon Intelligence gathering Tdl Favorable Monitor purchases of high explosives Study international relations Conduct reconnaissance activities Safeguards Td Favorable Inspect fuel cycle facilities Monitor material flows and inventories Identify loss of electrical generation Proliferation T Favorable Carry out diplomatic negotiations Type of Activity p production termination 2 Example of Activity Apply economic sanctions (a)Unfavorable - increases risk; favorable - decreases risk 43 weapons. Secondly, disregarding time efforts seriously alters results pro- duced on the relative risk of the once-through fuel cycle where spent fuel mounts up over a period of years to large quantities while radiation levels are continually decreasing. Based on these considerations, the preliminary results of the GA team are not discussed here. However, it should be point- ed out that the method outlined by GA could be applied successfully given a different set of assumptions were used for calculating their three factor formula. 44 Table VII Attribute Comparison Between Methods II III CHARM X = "Charm" factor Diverson Path TPWF = Total Proliferation Weight Factor SAI's (ERDA) Method no single factor produced 1. 1. Material Attractiveness l.A. Material Type l.B. Material Description l.C. Radiation Hazard 2. Distribution Parameter: Mass of Material Required for Explosive 3. Removal Mode Factor (single theft, substitution of inert material, substitution of isotopic material) I No. of Diversion Points 2. Self-Protection Factor (radioactivity level, etc.) 3. Minimum No. of Locations from which Material is Diverted 4. Effort to Process and Produce a Device from Stolen Material 5. Risk of Detection and the Risks Due to the Nature of the Material 1. Time Factors A. Time from Decision to First Weapon B. Time from Material Acquisition to First Weapon 2. Cost (to produce first weapon) 3. Professional Personnel Required 4. Material Unattractiveness A. Difficulty B. Detectability C. Interruptability D. Safeguardability IV V VI Selvaduray's Heuristic Method Safeguardability Index- Silvennoinen and Vira's Vulnerability Index Papazoglu et al.'s Multiattribute Approach 1. Mirimum Cost of Weapons Construction Once Material Available 2. Minimum Time to Weapon 3. Marginal Cost Incurred when Commercial Civil Program Amended to Weapons Purposes 4. Quality of Material for Weapons Purposes 5. Detectability of Weapons Production Attempt 6. Accessability and Accountability of Source Material 1. Weapon Development Time 2. Warning Period 3.A. Radioactivity Level ) Inherent B. Status of Information ) Difficulty C. Criticality Problems ) 4. Weapons Material Quality 5. Monetary Cost 1. Ability to Extract Pure Pu from Process 2. Location of Facility (on/off reactor site) 3. Decontamination Factor (radiation level) 4. Labor Intensity (No. of Personnel Required) 5. No. of ?rocess Streams VII Heising's Cost-Benefit Approach (Bayesian Decision Analysis) 1. 2. 3. 4. 5. 6. Domestic Availability of Technology Import Availability Capital, O&M Costs Suitability for Clandestine Operation Difficulty of Technical Implementation No. of Weapons Attainable from Material Flow 7. Quality of Weapons Material 8. No. of Technical Personnel Required 9. Level of Support Technology and Industry Required 10. Time Required to Construct Facility Fleming et al's Societal Risk Approach 1. Diversion Point 2. Mass (kg) 3. Characteristics of SNM at Time of Diversion a. b. c. d. Chemical Form Physical Form (solid, gas, etc.) Isotopic Composition Radiation Dose Level (low, medium or high) 4. Method of Conversion to Weapons Usable Material 5. Weapons Conversion Factor 6. Timely Warning Characteristics 45 III. Comparison of Methods/Conclusions It should be clear from this examination that all of the methods reviewed have used a common approach in developing quantitative methods by which to analyze nuclear fuel cycle diversion resistance. All methods have followed a process of attribute enumeration and definition, and have displayed a remarkable degree of agreement in this process. The attributes as defined by each of the methods are compared in Table VII. However, the degree of mathematical completeness evidenced in each method varies and is mostly dependent upon the analyst's knowledge of multi-attribute decision theory. For example, the work of Heising et al., Papazoglu et al. and Silvennoinen et al. all expressly use formal decision theory in setting up the mathematical relationships between attributes and final numerical indicators. Silvennoinen goes a step further into the realm of new mathematical techniques by inclusion of fuzzy integration. While Heising's method relies on a number of simplifying assumptions (i.e., linearity of utility functions and equivalence of Pareto weights), Papazoglu et al. follow through by use of Delphic techniques in establishing non-linear utility functions and non-equivalent weighting factors. These functions and factors are therefore based on expert opinion other than the analysts themselves and represent a concensus of those interviewed and questioned. * The work of SAI was not carried out to the point where a final quantitative ranking could be reached but helped in the process of *At least in the SAI unclassified work. Table VIII Correspondence Between Method Attributes VII Heising's ** Attributes 3 4 I CHARM 3 2 la-c/lb -- -- III SAI 3 2 4B 1C 2 1A -- -- 5 6 7 8 9 10 II Diversion Path (HEDL) -- 1B -- 4 -- IV Selvaduray Stanford * V VI Silvennoinen Papazoglu et al. Finland MIT 1,2 2 4 5 2 4A 3 5 3B/C 4D 4C 3 5 1 4 -- -- -- 1A/B -- -- VIII Fleming et al. GA -- 6 --- 3 4 3A 3A 1 -- 1 3 2,5 3 -- 4 6 * The correspondence shown is approximate as determined from the attribute definitions given in the papers (See Table VI). ** Heising's attributes are defined as follows (see Table VI): (3) Capital, O&M Costs, (4) Suitability for Clandestine Operation, (5) Difficulty of Technical Implementation, (6) No. of Weapons Attainable from Material Flow, (7) Quality of Weapons Material, (8) No. of Technical Personnel Required, (9) Level of Support Technology and Industry Required, and (10) Time Required to Construct Facility. -r_ 47 attribute definition (Papazoglu et al. derived many of their definitions Selvaduray's work, though not as well from the earlier SAI work). theoretically based in the methods of Operations Research, demonstrates the principle held by many decision theorists that decision theory is an accurate description of the logical human thought process. Selvaduray approaches the problem of choosing between alternatives by defining the relation: 11 r E j = R 11 10 E R i=l which can be placed into the notation of decision theory as follows: E =ErR 10-ER where i V(x ) x i i EX V Tl 1 V T E = Selvaduray's performance factor; r = Selvaduray's rating factor; R = Selvaduray's ranking factor; V(x ) = decision theoretic value function placed on attribute value x ; X = decision theoretic weighting factor placed on attribute i; and V = total value function summed over all attributes i for a particular technological route j. Thus, Selvaduray has managed to independently derive the same formulation for determining final evaluation indices as is done in formal decision theory. This result is of particular significance because it Table IX Final Results: Rankings Placed on Pathways to Weapons Usable Material (in %) Method Route to* Material 1 2 3 4 5 6 7 8 9 * VII Heising's Method I CHARM 22.5 21.7 6.6 4.5 6.6 10.6 9.0 7.0 11.0 7.1-21 10.4-43.5 0 0 2.6-6.3 27-42 5.4-6.9 9.3-34-9 0 II Diversion Path (HEDL) 6.1 25.3 1.6 3.6 15.7 15.7 15.7 15.7 .1 III SAI 0 IV Selvaduray Stanford V Silvennoinen Finland VI Papazoglu MIT 18.6 18.9 .4 .4 10.9 16.0 11.6 11.6 11.6 23.6 23.2 5.3 2.6 6.6 11.8 9.0 9.0 9.0 20.1-24.7 19.4-23.0 .8-3.0 .05-1.0 5.5-8.0 15.2-20.8 12.0-15.8 14.5-18.9 8.9-11.7 The routes to material are: (1) Research Reactor + Minimum Plutonium Recovery Plant (MPRP), (2) Production Reactor + MPRP, (3) Power Reactor + NPRP, (4) Power Reactor + Commercial Reprocessor, (5) Diffusion Cascade, (6) Centrifuge, (7) Aerodynamic Jet, (8) Electromagnetic Separation and (9) Accelerator. -A00 49 was work independently formulated apart from any specific knowledge of utility theory. However, Selvaduray's work fails to apply standard methods for arriving at value functions (see Appendix A) although he does use a questionnaire addressed to over 100 experts to determine the relative weighting factors X defined in his work. used to rank the eleven attributes The importance of Selvaduray's work is therefore less related to the efficacy of the heuristic method he applied than to the fact that his own careful study of the problem led to his "discovery" of utility theory, a sign that the theory does indeed model the logical human thought process. A Sample Problem Comparing Methods To gain further insight into how the various diversion resistance methods compare, it is useful to examine a sample problem. The problem examined is taken from the author's thesis in which nine currently available routes to weapons material were analyzed and compared with respect to their diversion resistance to obtain indicators of the relative probability that a non-weapons state would choose one route over another. Using the data of Table V, it is possible to place all methods on an equal footing to compare results. The approach taken is to determine the correspondence between the eight attributes defined in the author's work and the attributes defined in the other methods (TableVIII). Then, the other methods can be applied consistently to the available data. * Final results showing The method developed at GA by Fleming et al. was not found to be easily applicable to this problem without substantial use of computer codes unavailable to the author. Therefore, it was not included in the comparison conducted here. 50 relative rankings in % of the nine pathways to weapons usable material are given in Table IX. These results show a close agreement between the methods of Heising, Silvennoinen and Papazoglu et al. The Charm and diversion path results are significantly different but are probably not reliable because they are based on heuristic methods. Selvaduray's method is more firmly based in mathematical theory than are either Charm or diversion path methods, but again is not as firmly based as are those of Heising, Silvennoinen or Papazoglu. Note, however, that Selvaduray's results are not much out of agreement. To conclude, it appears that those methods based on utility theory render similar results. Method IV is probably less reliable than V, VI or VII since, although it bears a close resemblance to utility theory, it is not a formal application of the theory. In the author's estimation, methods I and II must be viewed very cautiously * because they are not based on any confirmed theoretical process. With regard to the work by Fleming et al. at General Atomic, a re-examination of the'definition used for consequences is suggested. The problem with defining proliferation risk to be proportional to number of weapons producible is that it does not reflect the concensus * An interesting result forthcoming from this exercise is that, given the validity of the data base, the commercial power reactor-commercial reprocessor route (#4) consistently is ranked as being very unattractive to a would-be proliferator, and is more diversion resistant than many of the independent pathways. However, although the relative risk appears small, it is not to be disregarded as insignificant. The proper conclusion to draw, as in the case of reactor accident scenarios, is that the risk does exist such that all reasonable technical precautions should be taken to minimize this risk. 51 that proliferation risk arises from the development of a first weapon independent from subsequent number produced. Further, since number of weapons produced is proportional to fuel cycle flow rate, an inherent bias is built-in against those fuel cycles which exhibit the highest nuclear material flow rates. Thus, more proliferation prone tech- nologies with smaller flow rates may be overlooked (e.g., reactor/MPRP route). the research This problem can be rectified if the method is recast to assess the risk as a function of the likelihood of a given fuel cycle contributing to the successful development of a first weapon capability. Finally, some comments are in order with respect to the proper interpretation of quantitative method results. results -- and methods -- peer review. It is important that be carefully scrutinized and subjected to As is the case with reactor safety analyses, this re- view process can reveal points of disagreement among experts, needed improvements in data/probability estimation in addition to other modifications. The proper role of analysis is to aid policy-makers in reaching reasoned conclusions; it is not a substitute for the policy-making process. As quantitative methods continue to develop, become more widespread and acceptable, the possibility for analysts to misuse analysis to support positions of advocacy will decrease as will the use of unstructured qualitative, intuitive approaches. The need for logically structured, technically based analyses is becoming greater as problems become ever more technical and complex. To dis- regard a scientific approach to difficult policy matters is a risk that implies far greater negative consequences than the risk that such methods will be applied incorrectly and/or without objectivity. 52 References 1. Nuclear Power: Issues and Choices, 1st ed., Ballinger, FordMitre Study, 1977. 2. Greenwood, T. et al., Nuclear Proliferation: Motivations, Capabilities and Strategy for Control, McGraw-Hill, New York, 1977. 3. President Carter, "Statement by the President on Nuclear Power", April 7, 1977. 4. Strauch, S., "Alternatives to Separation of Plutonium to Reduce Nuclear Proliferation Risk", Fuel Cycle Conference 1977, AIF, April 24-27, 1977, Kansas City. 5. Rasmussen, N.C. et al., Reactor Safety Study, WASH-1400, 1975. 6. Vaurio, J.K., "Response Surface Techniques Developed for Probabilistic Analysis of Accident Consequences", ANL, 1978. 7. Beck, W. and Schmidt, F.A., "Probabilistic Analysis of Codes and Calculations Results in Nuclear Reactor Safety", IFK-Stuttgart, FRG, 1978. 8. Webster, F.A. and Benjamin, J.R., "Probabilistic Analysis of Fire Risks for the Design of Fire Protection Systems in Nuclear Power Facilities", EDAC, USA, 1978. 9. Aldrich, D., Rasmussen, N.C. et al., "Examination of Off-Site Emergency Protective Measures for Core Melt Accidents", Sandia and MIT, 1978. 10. Candolfo, G. et al., "Some Aspects of the Risks Associated with a Mixed-Oxide Fuel Production Plant", AGIP-Nucleare and JRCIspra, Italy. 11. Fullwood, R.L., Ritzman, R. and Mendoza, Z., "Working Paper for a Probability/Consequence Analysis of a Nuclear Fuels Recovery and Recycling Center", SAI-054-77-PA, June- 1977. 12. Campbell, J.E. et al., "Development of Risk Assessment Methodology Applicable to Radioactive Waste Isolation", Sandia/USNRC, 1978. 13. Heckman, R.A., "Determination of Performance Criteria for HighLevel Solidified Nuclear Waste from the Commercial Nuclear Fuel Cycle: A Probabilistic Safety Analysis", Lawrence Livermore Laboratory, 1978. 14. Owen, P., "Analysis of the Risk from Sabotage of Nuclear Power Plants", Stanford University, EES Dept., Nov. 1975. 53 15. Topp, S.V., "Detailed Sabotage Analysis of a Commercial Reprocessing Plant", DPST-76-233, May 19, 1976 (classified). 16. Heinrich, L.A., "Safeguards Planning in a Plant Design Process", DP-MS-77-55, ANS San Francisco Meeting, Nov. 1977. 17. Heinrich, L.A., "Safeguards Requirements for a 10 MTU/day LWR Fuel Reprocessing Plant", DP STD-LWR-76-4, Nov. 16, 1976. 18. Levine, S., "Probabilistic Methods in the Nuclear Regulatory Process", USNRC, 1978. 19. Lovins, A., "Cost-Risk-Benefit Assessments in Energy Policy", George Washington Law Review, Vol. 45, August 1977, pp. 917-943. 20. Hoos, I., "The Assessment of Methodologies for Nuclear Waste Management", Transcript, U.C.Berkeley, 1976. 21. Taylor, V., "Subjectivity and Science: Belief", MIT Technology Review, 1979. 22. Personal communication with several engineers who shall remain anonymous here. 23. Albert, T.E. and Straker, E.A., "Analysis of the Proliferation Resistance of Alternative Fuel Cycles", Final Report to EPRI, A Correspondence about RP620-23, SAI-77-872-LJIF, Dec. 6, 1977. 24. D'Zmura, A.P., An Approach to Comparative Evaluation of Nuclear Fuel Cycle Proliferation Risk, Nov. 29, 1976. 25. Maltese, M.D.K., K.E. Goodwin, J.C. Schleter, Diversion Path Analysis Handbook, Vol. 1 - Methodology, Vol. 2 - Example, US ERDA, Division of Safeguards and Security, October 1976. 26. Kendrick, H. et al., A Preliminary Methodology for Evaluating The Proliferation Resistance of Alternative Nuclear Power Systems, SAI Report No. SAI-78-596-WA, June 15, 1977. 27. Selvaduray, G.S., "Comparative Evaluation of Nuclear Fuel Reprocessing Techniques for Advanced Fuel Cycle Concepts", Ph.D. Dissertation, Dept. of Applied Earth Sciences, Stanford University, Stanford, March 1978. 28. Silvennoinen, P. and Vira, J., "Quantitative Assessment of Relative Proliferation Risks from Nuclear Fuel Cycles", Technical Research Centre of Finland, Nuclear Engineering Laboratory, P. 0. Box 169, SF-00181, Helsinki 18, Finland. 29. Silvennoinen, P., Vieno, T., and Vira, J., "Fuel Cycle Optimization with Non-Proliferation Objectives", Transactions of the American Nuclear Society, Vol. 31, European Nuclear Conference, Hamburg, Germany, May 6-11, 1979, pp. 304-306. 54 30. "Viewpoints on Key Issues and Evaluations Criteria for Assessing the Potential of Alternative Nuclear Energy Systems for Improving Proliferation Resistance", TID-28078, Booz, Allen and Hamilton, Inc. (1977). 31. Saaty, T.L., "A Scaling Method for Priorities in Hierarchical Structures", J. Math. Psychol. Vol. 15, No. 234, 1977. 32. Heising, C.D., "The Reprocessing Decision: A Study in Policy Making Under Uncertainty", Dissertation, Dept. of Mechanical Engineering, Stanford University, 1978 (EPRI NP-931). 33. Keeney, R.L., Operations Research, Vol. 22, No. 1, 1974. 34. Sugeno, M., in Fuzzy Automata and Decision Processes, M.M. Gupta, Editor, North Holland Publishers, New York, 1977. 35. Papazoglu, I. et al., "A Methodology for the Assessment of the Proliferation Resistance of Nuclear Power Systems", MIT-EL-78-021, September 1978. 36. Fleming, K.N., Tully, G.R., and Deremer, R.K., "A Risk Assessment Methodology for Evaluating the Proliferation Resistance of Nuclear Energy Systems", GA-A15290, General Atomic Project 3769, Arms Control and Disarmament Agnecy Sponsorship, April 1979. A-1 Appendix A Examining Selvaduray's Method for Assessing the Safeguardability of Various Reprocessing Technologies Selvaduray defines five sub-attributes that influence the guardability of a given reprocessing method. safe- The way in which he assigns quantitative ratings (r) on the 48 possible combinations that could be assumed by interchanging possible attribute is Assigning ratings of between probably not mathematically justifiable. 0 to 1 based on equi-intervals of (.17) states units each, he constructs tables beginning with the Pu extractability of the process (yes/no), its location (on-site/off-site), decontamination factor (high, medium or low), labor intensity (high/low) and number of effluent streams (several/few). Through the rating assignment process, Selvaduray has implicitly assigned weighting factors (X) on the five sub-attributes as follows: X = .63 1. Pu separation A = 5 + 2. Location A = 2 +X 2 = .25 3. Decontamination factor A = .5 + 3 = .06 4. Labor Intensity A = .25 + 4 = .03 5. Effluent streams A = .17 5 = - .02 (A = mean difference in resulting ratings), ZA = 7.92 Had he placed the attributes in a different order, he would generate different A's for each attribute. Since he has not consciously as- signed the values noted, they probably do not adequately reflect his expert opinion (or anyone else's). Clearly, the implicit X's Selvaduray is using are not reflective of his own beliefs about the relative importance of each sub-attribute. Support Calculations for Table Appendix B IX I. CHARM Method X la Pathways 1 Z 3 4 5 6 7 8 9 Mass flow of Material (MT/yr) M MF N E H F PD si i D± lb Fraction Divertible w/ or w/o sfgds F 1-4 10-20 10-20 30 2-20 2-25 2-20 4-20 1 or less lc Simple Device Mass(MT) M 2 3 Self ProCost to tection Fabricate Factor (R/hr)Weapon(10 6 $) S P 100 100 105 105 1 1 1 1 105 10 10 1000 1000 1000 100 500 100 500 4 Time Required to Produce Device (yrs.) D 1-3 1-5 2-11 5-11 4-7 3-16 3-16 3-23 3-16 Charm X -4 3.3-40x10 2-20x10-3 9-18x10-9 2.7-6 xlO- 8 2.9-50x10-4 1.25-83x10-3 2.5-133x10-4 4.3-670x10-4 1.25-7x10-9 7.1-21 10.4-43.5 '.0 1%-0 2.6-6.3 27-43 5.4-6.9 9.3-34.9 Notes: Attribute (la) corresponds to the "Q" factor in the HEDL diversion path method, (lc) to the "M" factor and attribute (2) to the "RHF" (see part 11 of this appendix). Attribute (2) is rated the same way as for the diversion path method on a scale from 0.to 1 where 1 represents a very high level of self protection (radiation level v. high). Cost to fabricate weapon (factor P, attribute 3 here) is also rated on a scale from 0 to 1 where a very low cost corresponds to a very high Also, the time required to produce a weapon is rated on a 0 to I scale as was done in Silvennoinen's and attractiveness. Heising's work. MxF - Q in HEDL method. I-h B-2 HEDL Diversion Path Method II. (1) TPWF = MAF x DPF x RMF (2) MAF = MT? x MDF x RHF (3) DPF = /QIN~ Q - mass of material in fuel cycle, Ms - mass required for single device R.G. Pu 30 kg HE U-235 35 kg W.G. Pu 10 kg (4) WMF Path 1.0 0.75 0.1 Simple Theft Substitution of Inert Material Substitution of Isotopic Material Material Path Type MTF MDF RHF 1 2 3 4 5 6 7 8 9 R.G.Pu W.G.Pu R.G.Pu R.G.Pu H.E.U-235 H.E.U-235 H.E.U-235 H.E.U-235 R.G.Pu 1.0 1.0 1.0 1.0 .8 .8 .8 .8 1.0 .8 .8 .8 .8 .8 .8 .8 .8 .8 .7 .7 .1 .1 1.0 1.0 1.0 1.0 .1 (MT/yr) (MIr) Q M DPF RMF .03 .1 .1 .3 .105 .105 .105 .105 .005 .03 .01 .03 .03 .035 .035 .035 .035 .01 1 3.2 1.83 3.2 1.73 1.73 1.73 1.73 .71 0.75 1.0 0.75 1.0 1.0 1.0 1.0 1.0 1.0 MAF .56 .56 .08 .08 .64 .64 .64 .64 .08 TPWF % .43 1.79 .11 .256 1.11 1.11 1.11 1.11 .06 6.1 25.3 1.6 3.6 15.7 15.7 15.7 15.7 .1 E - 7.086 RMF (Removal Mode Factor): Simple theft of material is possible (5) only in case of commercial reprocessor where Pu is obtainable. Military routes require no diversion and therefore are not prone to be detected. Therefore, we modify the HEDL method here to include a category of material with the same rating as the simple theft path; that is, the path wherein material is derived from clandestine military operations outside of IAEA safeguards. MDF (Material Description Factor): In routes (1)-(4) and (9) Pu (6) nitrate solutions will need be handled in the PUREX NPRP and commercial reprocessor. In routes (5)-(8), binary compqunds, gases, etc., will need be handled in the enrichment processes. (7) 1(1: It is assumed uranium is enriched to 80% for enrichment routes. Q is defined as the mass of material that can be diverted without (8) detection during one year. III. SAI's Method SAI's method did not describe a way to convert qualitative/ quantitative data on attributes to quantitative single factors. Therefore, the SAI method is not applicable to the sample problem. B-3 IV. Selvaduray's Method Attributes L.Indep. U235JZEF 2 Location Pu Streams? (Weapons Material Covert Pathways ,ahwv Quality) Oaiv Yes 1 Research (High Quality) Reactor + , uitability Clandestine Operation) High 3 Decont. Factor (Opposite of Radiation Level High I I Rating 4 Labor Intensity (same as Stat, Of Information) V. Low 5 Effluent Streams (No. of_Weapons) Fev/ (10-r % (8.83) 1.17 18.6 (9.00) 1.00 18.9 AP 2 Production Reactor + MPRP Yes (Very High Quality) V. High High Med Several - 17) Power Reactor+- No Low Quality) MPRP Low V. Low High Several Low V. V. High Several 9.83 .4 ( .17) 9.83 .4 4 Commercial Reactor No (Low Quality) Low + Comm. Reprocessor _. ...... 5 Yes (Med. Quality) Low V. High High Several (5.17) 4.83 10.9 Centrifuge Yes (Med. Quality) High V. High Med Several (7.61) 2.39 16.0 Aerodynamic Jet Yes (Med. Med Several Diffusion Cascade 6 Quality) Low Y.High 4.48 11.6 8 Electromagnetic Separation Yes (Med. Quality) Low V. Low Med Several (5.52) 11.6 4.48 9 Accelerator Yes (Med. Low V. Low Med Several (5.52) 4.48 11.6 Quality) Z(1-r)-47.51 B-4 V. Silvennoinen Pathways et.al's K1 * 1 (C) Method (Finland) 1) 2 V(x 2 ) 3 V(X 3 ) V(X4 ) V(X 5) V(X 6 ) - V.Low 1.00 Low .75 High .75 High .75 .74 23.6 V.Low 1.00 Low .75 V.Low 1.00 High .75 V.High 1.00 .728 23.2 V.Low Cost 1.00 2 (M) - 6 5 4 ZXV(Xi) 3 (C) V.High Cost 0.00 - High .25 Med. .5 V.Low 0.00 Low .25 .165 5.3 4 (C) V.High 0.00 - High .25 V.High 0.00 V.Low 0.00 Low .25 .08 2.6 V.High 0.00 High .25 High .25 Low .25 Med. .5 .2075 6.6 - Med. .5 V.High 0.00 Med. .5 Med. .5 Med. .5 .37 - High .25 Med. .5 High .25 Low .25 Med. .5 .2825 9.0 - Med. .5 V.High 0-00 V.High 0.00 Low .25 Med. .5 .2825 9.0 High .25 Low .25 Med. .5 .2825 9.0 5 (M) 6 (M) 7 (M) 8 (M) 9 (M) - - Med. .5 High .25 *Silvennoinen et al's method characterizes a process by the source material available and the material flow rate achievable. Therefore, pathways here are described by source material type and expected flow rate to be consistent with the method here applied: (1) Res Reac + MPRP: Material Flow: R.G. Pu trom spent fuel, flow rate:<< 3OMTEM/yr, short cooling time. (2) Prod Reac + MPRP: Material -Flow: W.G. Pu from production fuel, flow rate: up to 30 MTHM/yr, short cooling time (3) Pow Reac + MPRP: Material Flow: R.G. Pu from spent flow rate: up to 30 MTHM/yr (probably less), cooling time to reduce radiation hazard (spent fuel assumed diverted from spent fuel ponds located in NWS that may be as old as 10 yrs.) (4) Pow Reac + Com Rep: Material Flow: R.G. Pu from com. PUREX plant process stream assumed located inside NWS, flow rate: up to 30 MTHM/yr (assumed standard size commercial reprocessor), radiation level low because of high decontamination factor C,10 6 for PUREX). (5) -(8) Enriehment Plants: clandestine plant. (9) Accelerator: Iaterial Flow: Material Flow: W.G. High Enriched U-235 passed through Pu; flowrate: up to 30kg Puf produced per year. The authors define two sets of weights, one for civilian(C) routes and X1 is assessed for M but not for C; X is assessed other for military(M). 2 for C but not for M: Unsep. R.G. Pu. X4 X3 X2 X Pu .30 .11 .17 .15 C .35 .20 .10 M The ratings are based on a 0 to 1 scale where 0 means the value to the non-weapons state if a path is least attractive and 1 most attractive: 0 .25 .5 .75 1 V. Low Attractiveness Low Med High V.High 11.8 IM. rapazoglu at. al.'s Method 3 Covert Pathways Research Reactor Development Time (Yrs.) Inherent Difficulity Warning Period Status of (%) Information Radiation Level (R/Hr ) 1-3 < 10 1-5 < 5 E(2,2) 10 2-11 V15 F(2,3) 10 * A(1,1) 10 2 Weapons Development Material Quality 6 (10 $) High R.G.Pu 10 Low V. High W.G.Pu 10 High Low R.G.Pu 1000 Criticality Low OeR? 2 Production Reactor + HPRP 3 Power Reactor + 5 U' 4 Power Reactor + Commercial Rep. 5 Diffusion* Cascade 6 Centrifuge 5-11 > 50 1(3,3) 105 High a%0 4-7 3-16 >30 > 15 H(2,3) 0 E(2,2) 0 High Med Low R.G.Pu 1000 Med HEU-235 1000 Med HEU-2 65 7 Aerodynamic Jet 8 Electro Magnetic 25-50 3 > 30 > 50 E(2,2) 0 E(2,2) 0 High High Med HEI-235 $I'-235 100 500 100 Separation * 9 Accelerator 25-50 __________________I C- Crisis Environment NC- Non- Crisis Environment > 30 E(2,2) I___________ 105 Med __________ Med HEU-235 500 VI. Papazoglu et.al.'s Attribute 1: (1) C '17NC -05- Method 1 -. 28 -1 Results of Papazoglu's Method: 2 -. 03 -. 07 Normalized Results In Right-Hand Column 10(1+E) Normalized 3c 4 5 -. 026 -. 026 0 0 0 0 0 0 -. 226-'336 -.146- -196 .774-.664 .854-.804 6.64-7.74 8.04-8.54 .19-20.8 .247-20.1 0 0 -.276- -406 -.176- -246 .724-.594 .824-.754 5.94-7.24 7.54-8.24 .t3-. #95 .23-.194 3a 3b 0 0 -. 02 -. 04 -. 06 -. 06 -. 026 -. 026 0 0 0 0 -. 04- .07 -.09- 715 -. 35 -.36 -. 154 -.154 -. 04 -.04 -. 01 -. 03 -. 04 -. 11 -. 884--.97 -. 864- -974 .116-.03 .136-.026 .3-1.16 .26-1.36 .009-.03 .008-.03 (4) C r31 NC -.12- -.13 -.07 -. 15 -.38 -. 38 -. 154 -. 154 -.04 -. 04 -.01 -. 03 -.04 -. 11 -1 -. 984--~994 0 .016-.006 0 .16-.06 0 - 0 .005-.001 (5) C .31 NC -11- -13 -.06- -07 -.13- -15 -.32 -.32 0 0 -.04 -.04 -.005 -.015 -.04 -.11 -.775- 7785 -. 725--.765 .225-.215 .275-.235 2.25-2.15 2.75-2.35 .07-.06 .08-.055 (6) C -28- -31 NC -. 1 - -13 -.095 -.098 -.06 -.06 0 0 -.02 -.02 -.005 -.015 0 -.02 .46-:49 -.323-~353 .54-.51 .677-.647 5.4-5.1 6.77-6.47 .157-.137 .208-.152 (7) C-31 NC--.13 -.06 -.13 -.06 -.06 0 0 -.04 -.04 -.005 -. 015 -.04 -.11 -.515 -.485 .485 .5 5 4.85 5.15 .14-.13 .158-.12 (8) C -28 NC i1 -.07 -.15 -.06 -.06 0 0 -.04 -.04 -. 005 -.015 0 -.02 -. 455 -.385 .545 -. 615 5.45 6.15 .159-.147 .189-.145 (9) C 731 13 NC -. -.06 -.13 -.06 -.06 -.02 -. 02 -.005 -. 015 -.04 -. 11 -.649 -. 619 3.51 3.81 .102-.094.117-.089 (2) C -17- ;3 NC-.05- -12 (3) C -. 25- -31 NC-.08- -.13 C-Crisis Environment NC-Non-Crisis Environment -.154 -.154 - .351 .381 0en VII. Heising's Method Applied to Problem Expressed with Papazoglu et. al.'s Attribute Definitions 1 Covert Development Pathways Time 3 Inherent Difficulty 2 Warning Period Research V. Low Low Reactor + .9 .7 Status of Information Radiation Level V. High 4 Quality Development Cost V. Low Weapons Materia: Critical'ft 5 Low Low High .9 .7 .7 .7 Med. Low Low .9 .x 1. yg 44l Normal -ized 5.5 .225 5.3 .217 1.6 .066 1.1 .045 MPRP 2 Production Reactor + MPRP 3 Power Reac- tor + HPRP Low .7 High .3 V. Low .5 .9 Med V.Low-Low .5 .2 .7 V. High .1 V. High V. Low .7 .9 .9 High Low .1 .3 .3 High Low V. High .1 .3 .1 High .1 Med. .5 V. High .1 1.6 .066 Med. .5 Med. .5 Med. .5 2.6 .106 High Med. High .1 .5 .3 2.2 .09 1.8 .07 2.7 .11 V. 4 Power Reac- tor + High .3 V. High V. Low .1 .1 V. High .1 Commercial toepo wer 5 Pathway Diffusion Cascade 6 Centifuge Pro Time.C Wann High .3 High .3 Low .3 V. High .1 Med. .5 Med. .5 -na- Med. -na- 7 Aerodynamic MAd Jet .5 8 Electro Magnetic High .3 .5 V. High .1 V. High .1 Med. .5 -na- High .1 Med. .5 Med. .5 Med. .5 High .3 Med. .5 V. High Med. .5 Med. .5 High .3 Separation 9 Accelerator .1 td