Proceedings of the 2008 Industrial Engineering Research Conference J. Fowler and S. Mason, eds.

Risk Analysis of Carbon Nanotube Production Processes under Uncertain Occupational Health Risks and Regulatory Standards
Zeynep D. Ok, Jacqueline A. Isaacs, James C. Benneyan Department of Mechanical & Industrial Engineering Northeastern University, Boston MA 02115, USA

Abstract
Despite the many touted benefits of nanotechnology, considerable uncertainty exists regarding full-scale production costs, occupational health risks, and future regulatory standards. Monte Carlo models were developed to investigate the implications of these uncertainties on high pressure carbon monoxide (HiPco) processes, a common single wall carbon nanotube (SWNT) production method. The health risks and production costs under alternate investments in occupational safety protection are shown to be significantly unpredictable, due to the large amount of uncertainty in model inputs. Experimental design methods therefore were used to identify which uncertainties have the greatest impact on this unpredictability to help manufacturers, policy makers, and researchers make informed decisions and research investments.

Keywords
Monte Carlo simulation, experimental design, SWNT, HiPco, occupational health

1. Introduction
Commercialization of nanotechnology is proceeding quickly in the global market due to its potential benefits in numerous application areas ranging from medicine to electronics. However, there remains a high degree of uncertainty surrounding the occupational, consumer, and environmental health and safety (EHS) risks of nanotechnology. Of the limited number of studies completed to date, several indicate that engineered nanomaterials may present potential risks to human health. Possible problems with these materials include their ability to penetrate dermal barriers, cross cell membranes, breach the gas exchange regions of the lung, travel from the lung throughout the body, and interact at the molecular level [1]. Studies on the toxicity of single wall carbon nanotubes (SWNTs) showed damage to lung tissue in mice [2-3], although further research is necessary to develop complete EHS information and consensus. Until more is definitively known about EHS risks of nanotechnology, manufacturers and policy makers are faced with significant uncertainty in how to proceed with commercialization of nanotechnology and safety requirements. Several risk analysis approaches historically have been used to provide decision making insight in various application areas involving uncertainty and risks. Apostolakis [4] discusses the use and benefits of quantitative risk assessment in decision making for complex technological systems, including: consideration of many different scenarios, development of a common understanding and systems understanding of the problem, increased probability of identifying complex interactions, and facilitated risk management. Monte Carlo (MC) simulation modeling is among the most widely used risk analysis approaches, with applications ranging from nuclear power to financial investment decisions [5-6]. In the present study, given the significant uncertainty in EHS information for HiPco SWNT manufacturing processes, MC models were developed to investigate and compare alternate worker health and safety protection investments, each with uncertain exposure levels, costs, and regulatory futures. These uncertainties in costs, exposure, and the level and timing of possible required standards are represented by probability distributions and chance events, and the impact of these factors was investigated on the expected values, variances, and probability distributions of production costs and occupational health exposure. Experimental designs were used to identify those factors and uncertainties that have the greatest impact on the variability of 5, 10, and 15year total occupational health (OH) risks and production costs and. Because the risks of commercializing SWNTs are likely to remain unclear for the foreseeable future, this approach can help to develop a more informed understanding of the range of possible manufacturing costs and OH risks and to identify optimal manufacturing decisions, regulatory policies, and research priorities.

1712

Ok, Isaacs, and Benneyan

2. Model Description
Life cycle assessment (LCA) and process-based technical cost models [7] were used previously to assess the economic and environmental impact of three established SWNT manufacturing processes: arc ablation, chemical vapor deposition (CVD), and high pressure carbon monoxide (HiPco). Given the high degree of uncertainty regarding fullscale manufacturing costs, OH risks, and probable future regulatory standards, MC models also were developed to explore their utility for predicting the potential range of long-term manufacturing costs [8]. Since the HiPco process resulted in significantly lower cost ($450/g versus $1,830/g for arc ablation and $1,586/g for CVD), risk models for HiPco processes were developed further [9], and the initial SWNT cost models were revised to include costs related to possible future EHS standards for worker protection. Four general levels of EHS standards (none, low, medium, and high) were defined in the model to represent the spectrum of possible regulatory futures. Assumptions for the types of worker protections at each EHS level are summarized in Table 1, based on discussions with experts in industrial hygiene and with costs estimated from other sources [9]. The HiPco SWNT process based technical cost model was re-designed using these cost estimates and assumptions to determine the expected per gram manufacturing costs. In actual practice, however, standards at any level or in any workplace are likely to have different and largely unknown implementation costs. These uncertainties are represented with triangular probability distributions, given the lack of much data, with the minimum and the maximum possible manufacturing costs under each EHS level initially set equal to 10% of their corresponding most likely values, as summarized in Table 2. Table 1: Summary of assumptions for four EHS standards levels
Level of EHS Standards None Engineering Controls General Exhaust - Ventilation Fume hoods Low 24 hr 1000 cfm ventilation rate, $10,000 capital cost, $3,000/year operating cost Medium 24 hr 1000 cfm ventilation rate, $10,000 capital cost, $3,000/year operating cost $4,000 capital cost for 6.25 ft2 equipment and $9,500 for 25 ft2 equipment* High 24 hr 1000 cfm ventilation rate, $10,000 capital cost, $3,000/year operating cost $4,000 capital cost for 6.25 ft2 equipment. $9,500 for 25 ft2 equipment* 50% decrease in labor productivity, 50% extra equipment cost 8 hours of training, $560/year instructor cost Biweekly monitoring, Equipment cost of $20,000 $950/worker/year 5 pairs/shift, $0.06/pair 5 pairs/shift, $0.09/pair 1/shift, $0.70 1/shift, $0.70 1 pair/ 30 hrs, $10/pair 1/shift, $4*** 5 pairs/shift, $0.09/pair

Enclosure of processes Annual worker training Air monitoring Medical monitoring 8 hours of training, $560/year instructor cost Monthly monitoring, equipment cost of $20,000 8 hours of training, $560/year instructor cost Weekly monitoring, equipment cost of $20,000

Admin. Controls

Personal Protective Equipment**

Gloves

Latex Nitrile Disposable HEPA filters

Respirators Tyvek suits

* Facility costs are not shown here, but are included in the cost model. ** Costs of hazardous waste disposal for personal protective equipment at the high level are not shown here but are included in the cost model. *** Costs of medical clearance and fit tests for respirators with HEPA filters are not shown here, but are included in the cost model.

Since no data are readily available regarding OH exposure for HiPco SWNT manufacturing process, an arbitrary 0 to 10 scale was assumed to represent the level of exposure, again with uncertainty represented via the triangular probability distributions shown in Table 2. For example, if a company does not adopt any standards, it is assumed that the minimum, most likely, and maximum annual values for worker exposure are 7, 9, and 10 units per year, respectively. However, if low levels of worker protection were implemented, then the minimum, most likely, and maximum exposure amounts would decrease to 6, 8, and 9 units. As the MC model executes, manufacturing costs and exposures under each level of EHS standards are randomly generated from these probability distributions. This logic is implemented in such a way to ensure that exposure and cost progressively decrease and increase, respectively, as a facility implements higher levels of worker protection.

1713

Ok, Isaacs, and Benneyan Table 2: Assumptions for EHS transition probabilities, costs, and exposures EHS Level (from) None Low Medium High Transition Probabilities (to) None .475 0 0 0 Low .475 .9 0 0 Med .025 .05 0.5 0 High .025 .05 .5 1 Cost Ranges [$/g] Min (-10%) $405 $414 $475 $594 Most Likely $450 $460 $528 $660 Max (+10%) $495 $506 $581 $726 Exposure Units (0-10 scale) Min 7 6 3 0 Most Likely 9 8 4 1 Max 10 9 6 2

Implementation rates of EHS standards also have significant uncertainty depending on technology, new research on health risks, regulations, and political forces. These likelihoods are defined by the transition probabilities also shown in Table 2, although different assumptions could be used in other analyses. Notationally, pii represents the probability of staying in a current level i and pij represents the probability of moving from level i to level j, with n, l, m, and h corresponding to none, low, medium, and high levels, respectively. By example, in this case the year-to-year likelihoods of remaining at no standards level and imposing low level standards are equally high (pnn= pnl = 0.475), whereas the probabilities of imposing medium or high levels of EHS standards are equally low (pnm = pnh = 0.025). If a low level of standards is imposed in some future year, then the probability of remaining at this low level of protection in subsequent years again will be higher (pll = 0.9) than of transitioning to medium or high level standards (plm = plh = 0.05), as shown in the second row of Table 2. If some medium level of standards is implemented, the probability of remaining at this level in the next year or moving to a high level both are pmm = pmh = 0.50, with this transition logic repeated annually for a 10 year analysis period.

3. Results
3.1 Joint Probability of Production Cost and Exposure Units Assuming the implementation rates given in Table 2, 10,000 MC replications of a 10-year analysis window produced a manufacturing cost expected value and standard deviation of $528/g and $65/g and an exposure mean and standard deviation 5.6 and 2.3 units, respectively. Note that in this type of analysis the expected value and standard deviation correspond to the most likely one-time outcome and the amount of uncertainty in this one-time outcome, respectively. That is, the actual manufacturing cost will lie anywhere between roughly $405/g and $726/g (a range of $321) and the actual exposure will lie anywhere between 0 and 10 units. These are very large ranges for manufacturing costs and exposure, due to the amount of uncertainty assumed in the model. Table 3: Joint probability distribution of HiPco manufacturing costs and exposure amounts
Ranges 400 < $ 425 425 < $ 450 450 < $ 475 475 < $ 500 500 < $ 525 525 < $ 550 550 < $ 575 575 < $ 600 600 < $ 625 625 < $ 650 650 < $ 675 675 < $ 700 700 < $ 725 725 < $ 750 Total 0<E1 0 0 0 0 0 0 0 0.002 0.006 0.016 0.014 0.007 0.002 0 0.047 1<E2 0 0 0 0 0 0 0.001 0.015 0.031 0.030 0.013 0.003 0 0 0.093 2<E3 0 0 0 0 0 0.001 0.014 0.039 0.037 0.013 0.001 0 0 0 0.106 3<E4 0 0 0 0 0.002 0.016 0.042 0.032 0.008 0.001 0 0 0 0 0.100 4<E5 0 0 0 0.003 0.018 0.045 0.023 0.004 0 0 0 0 0 0 0.092 5<E6 0 0.001 0.009 0.023 0.038 0.017 0.002 0 0 0 0 0 0 0 0.090 6<E7 0 0.018 0.086 0.087 0.017 0.002 0 0 0 0 0 0 0 0 0.210 7<E8 0.001 0.025 0.114 0.093 0.003 0 0 0 0 0 0 0 0 0 0.235 8<E9 0 0.004 0.014 0.009 0 0 0 0 0 0 0 0 0 0 0.028 9 < E 10 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.000 Total 0.001 0.048 0.223 0.214 0.078 0.081 0.082 0.091 0.082 0.060 0.029 0.010 0.002 0 1

1714

Ok, Isaacs, and Benneyan

10 9 8 7

Exposure [units]

6 5 4 3 2 1 0 $400 $450 $500 $550 $600 $650 $700 $750

Cost [$/g]

Figure 1: Joint probability density of HiPco manufacturing costs and exposure

Figure 2: Correlation of 10-year annualized cost and OH exposure (10,000 replications, r = -.945 (p = 0.000))

Figure 1 and Table 3 summarize the joint probability distribution obtained from the MC model of per gram manufacturing costs and exposure for a HiPco process, as well as the inherent tradeoff between production costs and workplace exposure. As illustration, the probability that production costs and exposure will simultaneously be between $450 to $475/g and 7 to 8 units, respectively, is 0.114, which is the largest shown joint probability volume. The marginal probability that the total cost (independent of exposure) will be less than $600/g is fairly high (roughly 80%), whereas the marginal probability that exposure (independent of cost) will exceed 6 is roughly 50%. However, note that the two measures are very highly correlated (Figure 2), with a correlation coefficient of r = -0.945 (p = 0.000), e.g., with low exposure linearly becoming less likely as production costs decrease. While the above results are based on a 10-year analysis window, Table 4 compares the means, standard deviations, and 95% confidence intervals under shorter and longer time horizons, decreasing or increasing expected costs and exposure in the obvious manners. Given the current significant uncertainty in EHS standards and health risks, similar analyses could be performed using different assumptions to provide further insights regarding worker protection, plant design, and technology investments. Table 4: Impact of analysis duration
Analysis Period 5 Year Cost [$/g] (Std Dev) Exposure [units] (Std Dev) $498 1.1 ($56) 6.6 0.04 (2) 10 Year $528 1.3 ($65) 5.5 0.05 (2.3) 15 Year $552 1.3 ($66) 4.7 0.05 (2.3) -1 0 1 5 10 15 Setting

Table 5: Experimental design settings

3.2 Characterization, Causes of Uncertainty, and Research Implications As the above results indicate, the current level of uncertainty in future EHS requirements, costs, and benefits translate to significant uncertainty in total long-term costs and occupational risks. Design of experiments (DOE) methods therefore were used to attempt to identify which variables and uncertainties (i.e., implementation probabilities and standard deviations of the cost and exposure of each EHS level) have the greatest impact on the unpredictability (i.e., variance) in exposure and production costs. Given the large full experimental space, two preliminary fractional factorial screening designs were run, using two different combinations of the input variables (length of the analysis time horizon, transition probabilities, and the mean and standard deviations in costs and exposure under each level of EHS standards) and using the high and low settings summarized in Table 5. In both sets of experiments, 10 results for four performance measure responses (cost mean, cost standard deviation, exposure mean, exposure standard deviation) were collected, with each combination of settings run for 10,000 replications and with 10 center points included in order to test for curvature. Table 6 summarizes the design settings, column assignments, confounding, and results for the first DOE, with standard deviation coefficients and p values

1715

Ok, Isaacs, and Benneyan computed via stepwise regression. The slight loss of balance and orthogonality was necessitated by the consistency logic mentioned in Section 2 to ensure within-replication transitions to higher standards yields increased costs and decreased exposure. Because the analysis window length (X1) which is the largest contributor to the cost and exposure means and variabilities is more of an analysis decision than an uncertainty, a second set of experiments were conducted omitting this variable. Table 7 summarizes these results, indicating now that uncertainty in EHS requirement probabilities is by far the biggest contributor to all four responses (mean and SD of cost and exposure) by a factor of 2 to 4, whereas the largest contributors to cost uncertainty are Cost, Low and to a lesser extent Cost, High, and Exposure, None is the largest contributor to exposure uncertainty. Interestingly, the effect of uncertain costs and exposures at medium levels of standards drops out of all models. Note that while the largest predictor (X5, implementation probabilities) is confounded with the 3-way None x Low x Medium interaction, this may be unlikely due to the insignificant Medium main effect. The implications of these results suggest that the greatest reductions in the uncertainty of occupational risks and production costs would result from focusing research efforts on developing better estimates first for the probabilities of future EHS requirements and then for the cost and risks under low and high standards, some of which might be more easily obtained than some of the other less significant and possibly less knowable factors. Table 6: First experimental design and results
Years Run 1 2 3 4 5 6 7 8 9 Cost mean p value Cost variance p value Exposure mean p value Expose variance p value X1 1 1 1 1 -1 -1 -1 -1 0 26.90 .00 755.00 .00 -.97 .00 .68 .01 Transition Probabilities X2 1 1 -1 -1 1 1 -1 -1 0 12.80 .00 411.00 .03 -.49 .00 .36 .09 Cost / Exposure Uncertainty Levels None X3 1 -1 1 -1 0.5 -1 0.5 -1 0 Low X4 (X1X3) 1 -1 1 -1 -0.5 1 -0.5 1 0 3.90 .04 Medium X5 (X2X3) 1 -1 -1 1 1 -1 -1 1 0 High X6 (X1X2X3) 1 -1 -1 1 -1 1 1 -1 0 E(Cost): V(Cost): -.12 .10 E(Expos): V(Expos): Y1 Mean Responses (n = 10) Cost Y2 Variance Exposure Y3 Mean Y4 Variance

563.26 4700.32 4.26 4.93 553.82 4707.44 4.63 5.72 536.24 3846.43 5.19 4.79 530.28 4519.68 5.56 5.44 500.79 3631.12 6.42 4.47 509.18 3357.45 6.20 4.46 475.29 2592.43 7.39 3.28 482.91 2150.82 7.37 3.20 528.29 4346.51 5.54 5.28 Y1 = 520 + 26.90X1 + 12.80X2 +3.90X4 Y2 = 3761 + 755X1 + 411X2 Y3 = 5.84 - .97X1 - .49X2 - .12*X4 Y4 = 4.62 + .68X1 + .36X2

4. Conclusions
Assessing the tradeoffs between manufacturing costs and occupational health consequences of nanotechnology production processes is especially difficult given the limited data on the health effects of nanoparticles. This paper illustrated the integrated use of Monte Carlo models and experimental design to study the impact of these uncertainties on the inability to predict long-term occupational health exposure risks, manufacturing costs, and inherent tradeoffs. Results underscore the observation that, given the extreme amount of uncertainty in the exposure risks and costs of different levels of protection, policy and manufacturing decisions should not be based on expected values alone. Experimental designs were used to identify which uncertain inputs have the greatest impact on the inability to accurately predict occupational health risks and production costs. Additional work is being conducted to further characterize what most affects the variability in both measures combined, such as in a total cost function or signal-to-noise ratios. Until research progresses on the EHS risks of nanotechnology, these types of analyses can provide useful information for private and regulatory decision-makers and for guiding research priorities.

1716

Ok, Isaacs, and Benneyan Table 7: Second experimental design and results
Cost / Exposure Uncertainty Levels None Runs 1 2 3 4 5 6 7 8 9 Cost mean p value Cost variance p value Exposure mean p value Expose variance p value X1 1 1 0.5 0.5 -1 -1 -1 -1 0 Low X2 1 1 -0.5 -0.5 1 1 -1 -1 0 4.25 .00 -117.00 .02 -.15 .01 -.17 .00 Medium X3 1 -1 1 -1 1 -1 1 -1 0 High X4 (X1X3) 1 -1 1 -1 -1 1 -1 1 0 Transition Probabilities X5 (X1X2X3) 1 -1 -1 1 -1 1 1 -1 0 13.87 .00 474.00 .00 -0.53 .00 0.41 .00 Y1 Mean 540.77 512.44 503.06 531.66 511.43 539.81 530.67 505.04 527.30 E(C): V(C): E(E): V(E): Responses (n = 10) Cost Y2 Variance 4628.08 3686.61 4022.29 4748.06 3460.55 4653.16 4760.82 3830.16 Exposure Y3 Y4 Mean Variance 5.02 6.04 6.41 5.35 6.28 5.12 5.46 6.46 5.15 4.59 4.81 5.71 4.72 5.54 5.89 4.92

4250.41 5.56 5.12 Y1 = 522.50 + 4.25X2 + 13.87X5 Y2 = 4227 - 117X2 + 60X4 + 474X5 Y3 = 5.75 - .06X1 - .15X2 - .53*X5 Y4=5.16.10X1.17X2.06X4+.41X5

60.00 .14

-.06 .13 -.10 .03

-0.06 .11

Acknowledgements
This work was supported by National Science Foundation grants SES-0404114 and EEC-0425826 through the Nanoscale Science and Engineering Center for High-rate Nanomanufacturing. The authors thank Dr. Michael Ellenbecker, Director of the Toxics Use Reduction Institute at the University of Massachusetts Lowell for his input on EHS assumptions used in this model.

References
1. 2. 3. 4. 5. 6. 7. 8. 9. National Institute for Occupational Safety and Health (NIOSH), 2007, Progress toward Safe Nanotechnology in the Workplace, Publication No. 2007-123:177. Lam CW, James JT, McCluskey R, Arepalli S, Hunter RL, 2006, A Review of Carbon Nanotube Toxicity and Assessment of Potential Occupational and Environmental Health Risks, Critical Reviews in Toxicology 36, 189217. Donaldson K, Aitken R, Tran L, Stone V, Duffin R, Forrest G, Alexander A, 2006, Carbon Nanotubes: A Review of Their Properties in Relation to Pulmonary Toxicology and Workplace Safety, Toxicological Sciences 92(1), 522. Apostolakis GE, 2004, How Useful Is Quantitative Risk Assessment? Risk Analysis 24(3), 515520. Pagani LP, Apostolakis GE, Hejzlar P, 2005, The impact of uncertainties on the performance of passive systems, Nuclear Technology, 149(2), 129-140. Kimura T, Shinohara T, 2006, Monte Carlo analysis of convertible bonds with reset clauses, European Journal of Operational Research, 168(2), 301-310. Isaacs JA, Tanwani A, Healy ML, 2006, Economic Assessment of SWNT Production, Proceedings of the 2006 IEEE International Symposium on Electronics and the Environment, 38-41. Ok ZD, Benneyan JC, Isaacs JA, 2007, Modeling Production Costs for SWNT Manufacturing Given Uncertain Health and Safety Standards, Proceedings of the 2007 IEEE International Symposium on Electronics and the Environment, 85-90. Ok ZD, Benneyan JC, Isaacs JA, 2008, Risk Analysis Modeling of Production Costs and Occupational Health Exposure of Single Wall Carbon Nanotube Manufacturing, Journal of Industrial Ecology: Special Issue on Nanotechnology and Industrial Ecology, accepted for publication.

1717

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.