Fumigation and Control Atmosphere

551

PS6-5 – 6156

Modeling the structural fumigation of flour mills and foodprocessing facilities

W. Chayaprasert1, D.E. Maier1, K.E. Ileleji1, J.Y. Murthy2

1 Graduate Research Associate, Department of Agricultural and Biological Engineering, Purdue University, 225 South

University Street, West Lafayette, Indiana, 47907 USA.2 Professor and Extension Engineer, Department of Agricultural and Biological Engineering, Purdue University, 225

South University Street, West Lafayette, Indiana, 47907 USA. Fax: (765) 496-1356, E-mail: maier@purdue.edu3 Assistant Professor and Extension Engineer, Department of Agricultural and Biological Engineering, Purdue University,

225 South University Street, West Lafayette, Indiana, 47907 USA4 Professor, Department of Mechanical Engineering, Purdue University, 585 Purdue Mall, West Lafayette, Indiana,

47907 USA

Abstract

3D computer models for structural fumigationwere developed based upon comprehensive datasets collected at a fumigation job in a commercialflour mill. The fumigation models were dividedinto two parts: internal and external flow models.The external flow model, which included theflour mill and surrounding structures, was usedto predict stagnation pressures on the mill’s wallsas a function of the wind speed and direction data.The pressure differences due to density differencesbetween the gas inside and the air outside of themill (stack effect) were also estimated using theenvironmental temperature and relative humiditydata. The combined effect of the stagnationpressure and the stack effect was used forestimating the boundary conditions of the internalflow model. The internal flow model incorporatedinterior details of the mill such as building plans,locations of major equipment, partitions, pipingand ducting. The idea of representing the cracksas an effective leakage zone was adopted. Theinternal flow model was able to yield a half-losstime (HLT) close to the HLT derived from thefield trial data. Therefore, it was concluded thatthe models were valid and the modeling

methodology established could be utilized forstructural fumigation modeling in any type ofstructure.

Key words: Structural Fumigation Modeling,Methyl Bromide Alternative, Sulfuryl Fluoride,Half-Loss Time, Computational Fluid Dynamics.

Introduction

The phase-out of methyl bromide (MB) as themajor fumigant for use in structural fumigationhas warranted the industry to seek for alternativepest control measures. However, fumigation withother MB alternatives is more costly and requiresa higher level of stewardship to be economicallycompetitive. Therefore, the key for successfuladoption of these alternatives lies in the efficiencyof its application during fumigation. Because itis not practical to perfectly seal the structure, thefumigation process can be better optimized onlyif the dynamics of gas movement in the fumigatedspace and the effects of environmental conditionson the process are well understood.

The decay of fumigant concentration is mainlydue to the transfer between the fresh air outside

9th International Working Conference on Stored Product Protection

552

and the fumigant-air mixture in the fumigatedspace. These movements are created by pressuredifferences across the building envelope. Hence,the analysis of fumigant gas leakage is similar tothat of infiltration through an air-tight building.Two main forces that create pressure differencesacross the building envelope driving naturalventilation and infiltration are the wind and stack(or buoyancy) effects (Ashrae, 2001). Thecombination of these two effects characterizeseach building and complicates the analysis. Thus,the Computational Fluid Dynamics (CFD)method is a proper method to solve this type ofproblem. In the heating, ventilation, and airconditioning (HVAC) industry, it has been usedto study wind-induced pressure on buildingsurfaces (Burnett et al., 2005; Senthooran et al.,2004) and contamination in building spaces(Cheong et al., 2003; Gilham et al., 2000; Sekharand Willem, 2004). Additional studies by otherresearchers have also been published on thesetopics. However, no published study has beenfond in the literature on the use of the CFDmethod for modeling the fumigation process inlarge structures.

This study was the first phase of an effort todevelop an analysis tool for structural fumigationand a precision fumigation controller that canautomatically apply and regulate fumigant gasduring fumigation. The primary objective of thisstudy was to develop flow models for predictionof fumigant leakage and distribution during thefumigation process in a reference flour mill. Thespecific objectives were to use a commercialCFD solver, Fluent® (Fluent Inc., Lebanon, NH):1) to construct a model of the flow outside thereference structure for predicting stagnationpressure profiles on the structure’s walls createdby prevailing wind, and 2) to construct a modelof the fumigation process that is able to reproduceconcentration data similar to those observedduring an actual fumigation job conducted in thereference structure.

The model development process was basedon the hypothesis that an acceptable modelshould be able to reproduce concentration datathat predicts HLT values similar to those

observed from experimental concentration data,given the same environmental conditions andfumigation practices. A fumigation experimentwas conducted as part of a commercial structuralfumigation in a grain processing facility in orderto collect experimental data necessary forvalidating the CFD models. The experiment andthe resulting data are described in the paperentitled Real-time Monitoring of a Flour MillFumigation with Sulfuryl Fluoride by the sameauthors.

Material and methods

External flow model

The domain of the external flow model wasset-up as a rectangular volume such that itincluded the grain processing building and thesurrounding structures. The perimeter boundariesfor the upwind side were specified as velocityinlets and the boundaries for the downwind sidewere specified as pressure outlets. To representthe atmospheric boundary layer, the velocityprofile at the inlet boundary(ies) is calculated as(Ashrae, 2001)

(1)

where UH is the local wind speed (m/s) atheight H (m) and Umet is the meteorological windspeed (m/s) measured at height Hmet (m). Thedescriptions and values for the other parameterscan be found in Ashrae (2001). Uniform staticpressure with a value of 0 Pascal was specifiedat the pressure outlet boundary(ies).

Results

Several steady-state flow simulations wereconducted with different fixed wind speeds anddirections. The wind directions were specified

Fumigation and Control Atmosphere

553

between 0 and 360 degrees with 15 degreeintervals. At each direction, six simulations wereconducted with six wind speeds: Umet = 1, 2, 4,6, 8, and 10 m/s.

Figure 1.a displays the resulting stagnationpressure profile when Umet was 4 m/s comingfrom a 30 degree angle with respect to the North.The experimental wind data however weremeasured at the grain processing building’s roofand for this particular case the wind speed at thebuilding’s roof was 4.1 m/s. The averagestagnation pressures on the north, east and westwalls were calculated separately using thepressure values and the wall areas shown in thefigure. These average pressures were then plottedagainst the wind speed calculated at thebuilding’s roof (Figure1.b). The averagepressures from the simulations with the otherboundary wind speeds (Umet = 1, 2, 6, 8, and 10m/s) were also plotted. It was found that therelationship between the roof wind speeds andthe stagnation pressures for each wall was asecond order polynomial fit (R2 � 1). Similarresults were found for the other wind directions.The coefficients of the fits were unique for everywind direction. Thus, at a fixed wind directionaverage stagnation pressures on the north, east

and west walls could be calculated for any roofwind speed.

Average stagnation pressures on the north, eastand west walls of the flour mill as a function ofwind speed. The values labeled 1, 2 and 3 areaverage pressure values corresponding to themagnified frames with the same numbers in (a).

The average stagnation pressures on the eastwall as a result of all external flow simulationswere comprehensively illustrated in Figure 2.These values were calculated using thepolynomial fits that were derived from theexternal flow simulation results. Similar chartswere also created for the north and west walls(data not shown). In general, regardless of winddirections, the stagnation pressures on all walls,when the wind speed was 1 and 2 m/s, weresubstantially lower (approximately within ±1Pascal) as compared to those for the higher windspeeds.

By mapping the plot in Figure 3 with theexperimental wind speeds and directions, theaverage stagnation pressures that would haveoccurred on the east wall during the fumigationperiod could be estimated. The averagestagnation pressures on the north and west wallscould be estimated using the same procedure.

Figure 1. Results of the external flow simulations when the wind direction was fixed at 30 degreeswith respect to the North. (a) Stagnation pressure contour on the building’s surfaces when Umet = 4m/s. The magnified frames labeled 1, 2 and 3 indicate the regions that were used to calculate averagepressures on the north, east and west walls, respectively. (b)

(a) (b)

9th International Working Conference on Stored Product Protection

554

Figure 3 displays time series of the resultingaverage pressures on the north, east and westwalls during the fumigation period. Most of thetime, positive pressures were exerted on the eastwall and the opposite occurred on the north andwest walls. The average pressures on the northwall were almost zero between the sixth andeleventh hours. Beginning at the eighteenth hour,the means of the average pressures on all wallswere predicted to decrease. These pressure series

were used as boundary conditions of the internalflow model which will be described in the nextsection.

Note that the pressures on the south wall ofthe flour mill were not taken into considerationbecause the south side of the flour mill is notexposed to the external environment. Instead, itis attached to the packaging building. During thefumigation, the walkway between the twostructures was sealed and thus it was assumedthat no gas movement occurred through the southwall of the mill.

Internal flow model

Unlike the external flow model, the dynamicsof fumigant concentrations indicates that the flowof a gas fumigant inside a structure is an unsteadystate problem. In the fumigation experiment, gasmovement between the fifth and sixth floors ofthe mill and between the mill and the otherconnecting structures were minimized by sealingthe air pathways between them. Assuming thatthese air pathways were perfectly sealed, thedomain of the internal flow model included onlythe first five floors of the mill. The domaincontained rectangular solid volumes representingmilling equipment such as rollermills, purifiers,sifters, pneumatic cyclones, tanks and temperingbins. The locations of the fumigant introductionsites and circulation fans in the model wereapproximately the same as the actual locationsdocumented from the experiment. The rate atwhich the fumigant (ProFume®) was released wascalculated using an estimated formula(Prabhakaran, 2005). A constant flow rate of 2.71m3/s was assumed for all circulation fans in themodel. Figure 4 shows the domain of the internalflow model. The total dimensions of the domainwere 26.5 m x 34.4 m x 27.6 m.

According to Ashrae (2001), the stackpressure difference for a horizontal leak at anyvertical location can be calculated as:

(2)

where ro and ri are the outdoor and indoor air

Figure 2. Average stagnation pressures on theeast wall as calculated using the polynomial fitsthat were derived from the external flowsimulation results.

Figure 3. Predicted average stagnation pressureson the north, east and west walls during thefumigation period. Elapsed time zero, 0, indicatesthe time of the first fumigant release.

Fumigation and Control Atmosphere

555

densities (kg/m3), respectively, g is the gravitationalconstant (9.81 m/s2), and H is the height above areference plane (m). HNPL is the height of the neutralpressure level (NPL) above the reference plane (m).In this study, HNPL was assumed to be the middleheight of the flow domain (13.8 m). Using Equation2, the equivalent pressure differences due to thestack effect were calculated along the height of theflour mill. The outdoor density was determined byapplying the psychrometric relationship to theexperimental data of barometric pressure,temperature, and RH. The indoor air density wassubstituted with the density of the fumigant-airmixture which was the mass integration of thefluid within the flow domain.

To model leakage, it was assumed that theleakage of fluid through all cracks on a wall canbe equivalently represented by the leakagethrough an equivalent leakage zone (ELZ). AnELZ was a pressure boundary condition with anassumed area (effective leakage area) and a losscoefficient. The pressure drop across the effectiveleakage area was calculated by:

(3)

where r is the fluid density (kg/m3), n is the

velocity through the vent (m/s), and kL is the losscoefficient. The effective leakage area for the ELZwas arbitrarily chosen as a circle with a diameterof 0.43 m. Five ELZs were placed on each of thenorth, east and west walls. Each ELZ waspositioned at the middle of the wall on each floor.As mentioned earlier, it was assumed that therewas no leakage through the south side of the milland therefore no ELZ was placed on the southwall. It was also assumed that the loss coefficients,kL, of all ELZs were identical. At each simulationtime step, the pressure value assigned to each ELZwas a summation of the average stagnationpressure, which was different for different walls,and the stack effect pressure, which varied withthe height of the ELZ. A time step of 3 minuteswas selected for the internal flow simulationsbecause it yielded a reasonable computer runtime.

Discussion

Equation 3 implies that kL is an indication ofthe gas-tightness of the mill envelope. It is aproperty of the mill and does not change withweather conditions. However, it is affected by thequality of the sealing method. With the pressureinputs having been determined from the externalflow simulations, the amount of leakage can becontrolled by adjusting the kL value of the ELZs.Several internal flow simulations were performedwith different kL values. At each simulation timestep, fumigant concentrations in the flow domainwere recorded at the same monitoring locationsas in the actual mill.

The simulation result that was closet to theexperimental concentration data is shown inFigure 5 which illustrates the simulatedconcentration curves of all monitoring points (18)in the first five floors when kL was equal to 90.The lower-case “m” in the labels indicates thatthese curves were simulated data. The primarydiscrepancies observed between the field trialdata and the simulation data were in the fumigantintroduction phase. In the simulation, there werefewer differences in the peak concentrations

Figure 4. The internal flow domain of the firstfive floors of the flour mill.

9th International Working Conference on Stored Product Protection

556

among the floors. Comparing floor by floor, theexperimental peak concentrations on the first andsecond floors were higher than the simulated peakvalues. The opposite occurred on the fourth andfifth floors. The simulated peak concentrations onthe third f loor were between the peakconcentrations measured by the two Fumiscopes.This resulted in much less time for uniform gasdistribution. The differences in the simulatedconcentrations at all locations were within 5 g/m3

at the fourth hour, while the same occurredapproximately at the sixth hour in the field trial.In addition, unlike the field trial data, the simulatedconcentrations at m5_1, m5_3, and m5_4 reachedthe equilibrium as quickly as those at the othermonitoring locations. These discrepancies were

anticipated mainly due to the simplification of theflow domain. Neglecting the fifth floor data, duringthe first fan-off period the fumigant in thesimulation was not as well mixed as in the fieldtrial. For this particular case, the greatestdifference between the highest (m1_1) and lowest(m4_2) concentrations was approximately 6 g/m3 or a difference of about 12 to 15%. Thisunderprediction of mixing was anticipated due tothe lack of modeling the convective currents inthe model. However, the underprediction was notconsidered critical because on average the modelwas able to yield a HLT value close to the HLTderived from the experimental data as discussedin the next paragraph. Once the second, third andfourth floor fans were restarted, the flow field wasdominated by momentum again and the fumigantquickly equilibrated. During the second fan-offperiod, the concentration differences within themill were still insignificant (within 3 g/m3

difference). The simulations with the other kL

values (data not shown) showed similarobservations, except that the peak concentrationsslightly decreased as the kL value decreased.

The average of all concentration curves inFigure 5 is plotted as the smooth solid curve inFigure 6a. When the data between the third andeighteenth hours of the average simulatedconcentration curve were fitted with the theoreticalconcentration decay equation (Equation 1 in theReal-time Monitoring of a Flour Mill Fumigationwith Sulfuryl Fluoride paper by the same authors),the simulated mill presented a HLT of 17.33 hours(R2 > 0.995), which was essentially identical tothe calculated 17-hour HLT of the actual mill. Thecurve with markers shown in Figure 6a indicatesaverage concentrations calculated using theexperimental readings. Comparing the twoaverage curves in Figure 6a, most of the time themodel slightly underpredicted the concentrationlevels. This resulted in the underprediction of theCt product as illustrated in Figure 6b. At the timeof unsealing, the Ct products (i.e., area under thecurve) of the experimental and simulated datawere approximately 950 and 850 g-hr/m3,respectively, or a difference of 10.5 %.

Figure 5. ProFume® concentrations obtainedfrom the internal flow model with kL = 90. The0th and 24th hours are the times when the firstProFume® introduction started and the mill wasunsealed, respectively. The first dash-dotted line(1.80) is when the second ProFume® introductionstarted. The second and fourth dash-dotted lines(3.51 and 14.50) are when the fans on the second,third and fourth floors were turned off. The thirddash-dotted line (9.05) indicates when these fanswere turned on.

Fumigation and Control Atmosphere

557

Figure 6. Comparison between the mill trial andsimulation results. (a) Average concentration plot.(b) Average Ct product plot.

Conclusions

This research presented a methodology forusing a commercial CFD solver, Fluent®, tomodel the fumigation process in a flour millbased on a data set collected at an actualfumigation site. The simulation illustrated severalminor discrepancies from the experimentalresults. However, the final achieved Ct productwas underpredicted by only 10.5 %. Furthermore,the HLT values derived from the internal flowmodel result and the field trial data wereessentially the same (17 hours). Therefore, theCFD models developed in this study were validand the established methodology could beutilized for fumigation process modeling in anytype of structure.

Given weather conditions, the models can beused to predict fumigation characteristics suchas fumigant movement paths, concentrationdistributions, and leakage rate. The effects offumigation variables such as wind speed anddirection, capacity and placement of circulationfans, and fumigant release time on the efficacy ofthe fumigation process are currently beingevaluated. The results from the simulations willprovide insight into understanding the dynamicsof the structural fumigation process and helpfumigators to correctly determine the amount offumigant to be used, which in turn will yieldincreased efficacy and more successfulfumigation jobs.

Acknowledgments

This study was funded by the USDA-CSREES Methyl Bromide Transition Programunder project grant 2004-51102-02199“Fumigation Modeling, Monitoring and Controlfor Precision Fumigation of Flour Mill and FoodProcessing Structures.” The cooperation, inputand help of Mr. John Mueller, Mr. David Mueller,Mr. Peter Mueller and the staff of FumigationService & Supply Inc., Indianapolis, Indiana hasbeen greatly appreciated throughout this project.The cooperation of Dr. Suresh Prabhakaran, Mr.Marty Morgan and other staff at DowAgroSciences, Indianapolis, Indiana, as well asthe staff of several flour mills is alsoacknowledged.

References

Ashrae, 2001. ASHRAE Handbook -Fundamentals. Atlanta, GA: AmericanSociety of Heating, Refrigerating and Air-Conditioning Engineers, Inc.

Burnett, J., Bojic, M., Yik, F., 2005. Wind-induced pressure at external surfaces of ahigh-rise residential building in Hong Kong.Building and Environment 40, 765-777.

(a)

(b)

9th International Working Conference on Stored Product Protection

558

Cheong, K.W.D., Djunaedy, E., Poh, T.K.,Tham, K.W., Sekhar, S.C., Wong, N.H.,Ullah, M.B., 2003. Measurements andcomputations of contaminant’s distribution inan office environment. Building andEnvironment 38, 135-145.

Gilham, S., Deaves, D.M., Woodburn, P.,2000. Mitigation of dense gas releaseswithin buildings: validation of CFDmodelling. Journal of Hazardous Materials71, 193-218.

Sekhar, S.C., Willem, H.C., 2004. Impact of

airflow profile on indoor air quality - a tropicalstudy. Building and Environment 39, 255-266.

Senthooran, S., Lee, D.-D., Parameswaran, S.,2004. A computational model to calculatethe flow-induced pressure fluctuations onbuildings. Journal of Wind Engineering andIndustrial Aerodynamics 92, 1137-1145.

Prabhakaran, S., 2005. Personalcommunication with Dr. SureshPrabhakaran, Bio/Technical Expert, DowAgroSciences, Indianapolis, Indiana. 20June 2005.