United States
[Environmental Protection
Agency
Risk Reduction
Engineering Laboratory
Cincinnati, OH 45268
Research and Development
EPA/600/S2-90/028 Sept. 1990
Project  Summary
Refinement  of  a  Model  to Predict
the  Chemical  Permeation of
Protective  Clothing  Materials
Rosemary Goydan, Todd R. Carroll, Arthur D. Schwope, and Robert
C. Reid
   A prototype of a predictive model for
estimating chemical permeation through
protective clothing materials was refined
and tested. The model applies Fickian
diffusion theory and predicts permeation
rates and cumulative permeation as a
function of time for five materials: butyl
rubber, low density polyethylene (LDPE),
natural rubber, neoprene, and nitrile
rubber. The model provides two ap-
proaches to estimate the solubility, one
using  a  group contribution approach
(UNIFAP S) and the second using an
equation of state approach (EOS S). The
model provides one approach to esti-
mate the diffusion coefficient (CORR D).
Refinement of the model was investi-
gated through a preliminary analysis of
the concentration dependence of the
diffusion coefficient. A finite difference
technique was developed and, for 50%
of the cases analyzed, the permea-
tion-time behavior could be described
more accurately assuming concentra-
tion dependence. No correlation, how-
ever, was identified to apply this finding
in a predictive mode. Correlations devel-
oped previously to estimate constant D
values (CORR D) were refined using a
larger data set. The accuracy and limita-
tions of the refined model were evalu-
ated by comparing model predictions
with literature data. Overall, the accu-
racy of the model is fair; for 200 data sets
representing a range of chemical types,
70%-80% of the predicted permeation
rates were within an order of magnitude
of the measured values. The UNIFAP S/
CORR D approach was more accurate
than the EOS S/CORR D approach, how-
ever the former could not be applied in
many  cases because UNIFAP param-
eters are not available for all functional
groups.
   This Project Summary was developed
by EPA's Risk Reduction Engineering
Laboratory, Cincinnati, OH, to announce
key findings of the research report that
is fully documented in a separate report
of the same title (see Project Report
ordering information at back).

Introduction
   Section 5 of the Toxic Substances Con-
trol Act requires prospective manufacturers
to submit Premanufacture Notifications
(PMNs), which are reviewed by the EPA's
Office of Toxic Substances (OTS), before
manufacturing or importing new chemicals.
A primary objective of the review is to assess
the  potential  risks to  human  health that
could result from dermal or inhalation ex-
posures during the manufacture,  process-
ing, or end use  of the  PMN substance. In
those cases in  which  the  PMN submitter
recommends protective clothing as a way to
minimize dermal exposures, OTS needs a
rapid and well-substantiated method to as-
sess the ability of protective clothing to act
as a barrier to the PMN chemical.
   The EPA's Office of Research and De-
velopment, in support of OTS, has explored
approaches to develop predictive models
and test methods for estimating the barrier
properties of protective clothing materials.
Chemical permeation of clothing materials
has been the focus because it is an impor-
tant mechanism by which chemicals can
cross through protective clothing.  Previous
reports prepared under this ongoing effort

-------
describe a prototype of a predictive model
for estimating  permeation  resistance and
guidelines for  specifying and interpreting
results from chemical resistance tests. This
project report summarizes recent efforts to
refine and test the prototype model.
    The following criteria were specified by
OTS to  guide  model development.  The
model should:
     be easy to use,
     apply to a wide range of chemical and
     protective clothing materials,
     apply to new chemicals not used to
     develop the model,
     not require data other than those
     typically supplied in PMN submissions,
     predict the  cumulative mass of
     chemical that permeates the clothing
     material as a function of time, and
     enable prediction of  breakthrough
     times at specified permeation rates or
     cumulative amounts permeated.
Accuracy requirements were net  specifi-
cally defined.
    The prototype  model  uses diffusion
theory  and  Pick's law to  estimate neat
chemical permeation as a function of time
through five clothing materials: butyl rubber,
LDPE,  natural rubber, neoprene, and nitrile
rubber. The Pick's law approach was se-
lected because it was judged to provide the
best opportunity to satisfy the above crite-
ria. While other methods (e.g., statistical
correlation  methods) may demonstrate
better  accuracy  in some cases, such
methods often cannot predict permeation
behavior as a function of time and  lack
sufficient theoretical basis for extrapolation
to new chemicals. The prototype  model
estimates two parameters: the solubility (S)
and the  diffusion  coefficient (D) of the
chemical in the protective clothing material.
Two approaches to estimate S are  pro-
vided, one using a group contribution ap-
proach (UNIFAP S) and the second using
an equation of state approach (EOS S). The
prototype provides one approach  to esti-
mate D (CORR D). The prototype  model
was coded in FORTRAN, runs on a. personal
computer, is easy to use, requires minimal
data inputs, and predicts  permeation be-
havior as a function of time including break-
through times. This report summarizes the
results of a  study to refine the prototype
model and document its accuracy and limi-
tations.

Refinement and Testing  of
Predictive Model
   Our previous validation of the prototype
model was limited by the availability of reli-
able data. Consequently, the priorities of
this effort were  (1) to  identify  additional
permeation data available for model devel-
opment and testing, (2) to refine the proce-
dure to estimate D, and (3) to test the refined
permeation estimation model to document
its accuracy and limitations.
   Permeation model refinement and test-
ing were performed through the analysis of
an additional, large set of permeation data
that had been generated for the National
Toxicology Program (NTP) by the Radian
Corporation. The data set includes perme-
ation data for approximately 50 chemicals
per material for butyl rubber, natural rubber,
neoprene, and  nitrile  rubber.  Permea-
tion-time curves were made available to us
for a total  of approximately 40 of the 200
chemical/material combinations. For the
remainder of the data set,  only undefined
breakthrough times and  steady-state per-
meation rate values were available. Model
refinements focused on  improving proce-
dures to estimate values for the diffusion
coefficient. No revision was made to either
of the solubility estimation techniques. The
D refinement  effort was twofold: (I) to un-
dertake a  preliminary investigation of the
importance of the concentration dependence
of D to accurately predict permeation be-
havior as  a function of  time, and  (2) to
explore approaches for improving the esti-
mation of constant D values using the NTP
data set.

Concentration Dependent Diffusion
Coefficients

   The investigation of concentration de-
pendent diffusion coefficients was explor-
atory and  used  numerical methods to cal-
culate permeation-time profiles for general
cases of concentration dependent behav-
ior. Two general cases were studied, D as
a linear function and as an  exponential
function of concentration, in addition to the
constant D case.
   A finite difference numerical technique
was  developed and used  to calculate  a
parametric series of permeation-time profiles
for these functions. The permeation-time
profiles were  plotted on a dimensionless
scale so that their characteristic curve shape
could be compared. In general, concentra-
tion dependent diffusion coefficients reduce
the initial rate of permeation (i.e., increase
the breakthrough  time) and produce  a
sharper increase,  at later times,  to  the
steady-state rate compared with the constant
D case. The parametric profile curves were
then  used  to analyze for the presence of
concentration dependent behavior for 31
chemical/polymer combinations. Overall,
50% of the chemical/polymer combinations
tested  show some concentration depen-
dent  D behavior. Because no correlation
was  identified  to apply  this  finding  in  a
predictive mode, however, the permeation
estimation model will continue to be based
on the assumption of a constant D.

Estimation of Constant Diffusion
Coefficients

   As an extension of our previous efforts,
we again  investigated correlations of the
constant D values calculated  from the
available permeation data set with charac-
teristics representative of the permeant size
and shape. These properties included mo-
lecular weight, molecular connectivity, sur-
face/volume ratios, and the acentric factor.
As found previously, the best correlation
was  with  molecular weight. Correlations
with  other  properties were less accurate
and imposed additional estimation require-
ments on the user. Revised correlation
equations were established forbutyl rubber,
natural rubber, neoprene, and nitrile rubbers.

Permeation Model Testing
   The refined permeation estimation model
was tested by comparing the model predic-
tions  with permeation  data from the NTP
data set.  For solubility estimation, both the
Oishi and Prausnitz UNIFAP group contri-
bution technique (UNIFAP S) and the Kumar
equation  of state technique (EOS S)  were
used. For D estimation, the revised equations
for estimating constant D values from the
permeant molecular weight were  used
(CORR D).
   Model predictions using the UNIFAP S/
CORR D approach require  input of the
following  properties for  the permeant of
interest:
      Molecular weight, g/mol,
      Liquid density, g/cm3, and
 •    Chemical structure, defined accord-
      ing  to a specified  set of functional
      group designations.
   Model predictions  using the EOS  S/
CORK D  approach require input of the fol-
lowing permeant properties:
      Molecular weight, g/mol,
      Liquid density, g/cm3, and
      Vapor pressure, mm Hg.
   Both approaches require the user to
select the protective clothing material of
interest and input the material thickness.
   For each approach, the permeation rate
and the cumulative mass permeated as a
function of time were predicted. From these
curves, breakthrough times and steady-state
permeation rates are  calculated.  Break-
through times were estimated on the basis
of thefirstreportedrateorcumulative amount
permeated. The  permeation data  subset
with values reported as a function of time,
36 chemical/material combinations,  was
used to judge the accuracy of the models in
predicting the full permeation-time curve.
Because of the limited permeation-time data

-------
set available, a second validation effort
was undertaken using the remainderof the
NTP data for which only undefined break-
through times and steady-state rate values
were reported, approximately 200 chemi-
cal/material  combinations.   Since
steady-state permeation  rates are inde-
pendent of test conditions, these values
are useful for model validation purposes.

Prediction Accuracy

   The  model predictions by both ap-
proaches were compared with the mea-
sured data to assess the prediction accu-
racy. Accuracy was judged by calculating a
percent error for each predicted value:
   Percent error (%)  =  [(Predicted
   value-Measured valuej/Measured
   value] * 100
   Overall, the accuracy of the model is
fair for predicting the permeation of organic
chemicals through butyl rubber,  natural
rubber, neoprene, and nitrile  rubber The
model prediction  results are summarized
in Tables 1 and 2 in terms of the number of
predictions that fall within specific (although
arbitrary) accuracy ranges defined using the
percent error values. Table  1 summarizes
the results for the prediction of steady-state
permeation rates and Table  2 summarizes
the results for breakthrough time prediction.
The results are organized by modeling ap-
proach (UNIFAP S/CORR D versus EOS S/
CORR D) and by polymer type.
   The accuracy of  the permeation model
predictions was  also  analyzed  using  an
analysis of  variance (ANOVA)  statistical
technique. Only the steady-state permeation
rate prediction results were  analyzed. The
analysis shows that  the model is not suffi-
cently "accurate" inthetraditionalsensethat
rigorous statistical descriptions of accuracy
are not meaningful. Consequently,  our dis-
cjssion of model accuracy  involves only
qualitative observations.
   When applicable, the UNIFAP S/CORR
C modeling approach was found to be more
accurate than the EOS S/CORR D approach.
As noted in Table 1, the UNIFAP S/CORR D
approach was used  to predict steady-state
permeation rates for a total of 122 chemical/
material combinations for the four clothing
materials.  There were, however, an addi-
tional 96 chemical/material combinations
that could not be addressed because the
required UNIFAP parameters are not avail-
able for several, common  chemical func-
tional groups. Seventy-five percent of these
combinations involved neoprene or nitrile
rubber.
   For the cases in which the UNIFAP S/
CORR D approach could be applied, the
predict ed values overall were within afactor
of 5 (i.e., percent errors in  the range from
-80% to +400%) of the measured values for
64% of these combinations and within a
factor of 10 (i.e., percent errors in the range
from -90% to +900%) for 80% of the combi-
nations. The predictions for natural rubber
and  nitrile rubber were  somewhat more
accurate than those for butyl  rubber and
neoprene. Also, the model tended to over-
estimate the steady-state permeation rates
for neoprene and nitrile rubber.
 Table 1. Summary of Model Accuracy for Predicting Steady-State Permeation Rates
                                               No. of Predictions in Accuracy Range

UNIFAP S/
CORRD
Butyl
Rubber
Natural
Rubber
Neo-
prene
Nitrile
Rubber
EOSS/
CORRD
Butyl
Rubber
Natural
Rubber
Neo-
prene
Nitrile
Rubber
No. of
Chem-
icals in
Data Set


58
47

57

56



54
37

54

50

No. of
Chem-
icals Model
Applied


44
36

21

21



54
37

54

50

<-90%
(900%
(>factor of
10)


5
3

6

4



3
6

7

2

Predic-
tions Within
-80% to 400%
Error
(factor of ±5)


57%
70%

62%

71%



57%
43%

35%

62%

Predic-
tions Within
-90% to 900%
Error
(factor of ±10)


80%
81%

71%

81%



69%
65%

52%

78%


-------
Table 2.  Summary of Model Accuracy lor Predicting Breakthrough Times

                                                No. of Predictions in Accuracy Range

UNIFAP S/
CORRD
Butyl
Rubber
Natural
Rubber
Neo-
prene
Nitrite
Rubber
EOSS/
CORRD
Butyl
Rubber
Natural
Rubber
Neo-
prene
Nitrite
Rubber
No. of
Chem-
icals in
Data
Set


10

5

11

10



1O

5

11

10

No. of
Chem-
icals
Model
Applied


7

4

4

6



1O

5

11

10

<-9O%
(<1actor
of -1O)


2

1

1

2



2

1

2

2

-9O% to
-80%
(factor of
-5 to -1O)


2

1

O

2



3

1

0

1

-8O% to
4OO%
(factor of
± 5)


3

2

3

1



4

3

5

3

4OO% to
9OO%
(factor of
5 to 1O)


O

O

O

1



1

O

1

2

>9OO%
(>factor
of 1O)


O

O

O

O



O

O

3

2

Predic-
tions Within
-80% to
400% Error
(factor of ±5)


21%

5O%

75%

17%



4O%

60%

45%

3O%

Predic-
tions Within
-9O% to 9OO%
Error
(factor of± 1O)


71%

75%

75%

67%



8O%

8O%

55%

60%

   Table 2 reports the results forthe UNIFAP
S/CORR D prediction of breakthrough times
forthe 21 of 36 chemical/material combina-
tions that could be addressed  using this
approach.  The predicted breakthrough
times, used to judge model accuracy  at
early times, were within a factor of 5 of the
measured  values for only 43% of these
combinations and within a factor of 10 for
71% of  the combinations on average.  In
general  the method  underestimated the
measured values.
   In  comparison, the  EOS S/CORR  D
could be used to make predictions for all of
the chemical/polymer combinations in the
data set, however,  the accuracy  of this
approach is only fair to poor. The predicted
values for steady-state permeation rates
were within a factor of 5 of the  measured
values for only 50% of these combinations
and within  a factor of 10 for 66% of the
combinations. Predictions for  neoprene
rubber were poor. Similarly, as summarized
in Table  2, the predicted breakthrough time
values were within  a factor of 5 of the
measured values for 42% of these combi-
nations and within a factor of 10  for 67% of
the combinations. In general, this method
underestimates the breakthrough times and
the steady-state permeation  rates.

Applicability and Limitations

   The  permeation estimation model  is
based on the assumption of ideal Pick's law
behavior. The Pick's law equations used to
develop the model assume:
     continuous  contact of the liquid
     chemical over the entire surface area
     of the polymer for the duration of the
     exposure. The model does not address
     intermittent or splash exposures.
     no external phase resistances exist at
     the downstream clothing material sur-
     face (i.e.,  the inside surface of  the
     clothing material that would contact
     the skin during clothing use). This
     assumption may not be appropriate
     when actual  use conditions are con-
     sidered.
     the diffusion  coefficient and clothing
     material thickness are constants. Ef-
     fects such as polymer swelling and
     structural relaxations, which produce
     concentration   dependent  and
     non-Pickian diffusion behavior, are not
     addressed.
   The model does not address multicom-
ponent solutions and was developed  to
treat only five, generic, "homogeneous" or
"isotropic" protective clothing polymers: butyl
rubber,  natural  rubber,  neoprene,  nitrile
rubber,  and LDPE. The  model  currently
does not address polymer blends or lami-
nates. Also, the solubility and diffusion coef-
ficient estimation techniques apply  only at
or near 25°C.
   As  noted above, the UNIFAP S tech-
nique has the critical limitation that it cannot
be applied to many chemical/polymer com-
binations in  its present state of develop-
ment. Whereas the EOS S/CORR D ap-
proaches require only basic physical prop-
erties of the permeant as input, the Oishi
and Prausnitz UNIFAP group contribution
technique to estimate S requires specific
functional group designations and group
parameters as input. Thus, the applicability
of the approach is limited by the available
functional group parameters. The applica-
bility is fairly broad for butyl rubber, LDPE,
and  natural rubber but very  limited for
neoprene and nitrile rubber.

Conclusions and
Recommendations
   The predictive model evaluated in this
study can be used to estimate permeation
behavior,  however, the  accuracy of the
model is only fair when tested using a range
of organic chemicals and clothing materi-
als. Further  improvement of model accu-
racy is not possible unless a larger set of
well-documented permeation  data as a
function of time  are available  for model
refinement.
   The assumption of a constant D  may
reduce the accuracy of the permeation pre-
dictions because, in 50% of the cases ana-
lyzed, we found that the permeation behav-
ior as a function of time was more accurately
described using a concentration dependent
D. The permeation-time  data set  was too
small, however, to develop predictive corre-

-------
lations. The correlations developed previ-
ously to estimate constant D values were
refined by analyzing a larger set of "aver-
age" constant D values compiled under this
effort.
   Overall, the accuracy of the refined model
isfairforpredicting the permeation of organic
chemicalsthrough butyl rubber, natural rub-
ber, neoprene, and nitrile rubber. The model
predictions were often within the range of
experimental values reported in cases where
multiple or replicate tests were performed.
The UNIFAP S/CORR D modeling approach
was more accurate than the EOS S/CORR
D approach, however the UNIFAP S ap-
proach could not be applied in all cases.
Other limitations of the model are that it
assumes continuous contact of the clothing
material with the permeating chemical, it
does not address chemical mixtures, and it
applies only at  25°C.
   Ourprimary recommendation isthatOTS
should reevaluate  its  requirements for a
permeation prediction model,  specifically
regarding prediction accuracy, before fur-
ther efforts are undertaken  for predictive
model development. Acceptable model ac-
curacy  must  be defined and prioritized
relative to broad applicability, ease of use,
and cost of development.
   At present, OTS can use the permeation
model to estimate  PMN chemical perme-
ation of protective clothing materials so long
as they  recognize the accuracy and appli-
cability limitations of the model. The model
is only useful for order of magnitude esti-
mates. When applicable, we recommend
using the UNIFAP S/CORR D approach.
The EOS S/CORR D approach should be
used only for cases in which the UNIFAP S/
CORR D approach cannot be  applied be-
cause the required group parameters are
not available.  In any  case, the  EOS S/
CORR D approach is not recommended for
predicting permeation through neoprene
materials because of the poor  accuracy of
the predictions.
   No efforts to improve the present model's
accuracy should be undertaken until a larger
set of well documented  permeaition-time
data can be obtained. OTS should contact
the NTP and other researchers to obtain
additional, existing permeation-time  data
for model refinement and validation. If such
data can be obtained, our specific recom-
mendations are:
     to  test the accuracy and applicability
     of the present model if the data are for
     LDPE.
     to  consider expanding the model to
     other clothing materials if the data are
     for materials not included in the present
     model.
     to  pursue approaches  to predict the
     concentration dependence of the dif-
     fusion coefficient.
     to expand the range of applicability of
     the UNIFAP group contribution ap-
     proach to  predicting S. One option
     would be for OTS to support research
     to generate the required parameters
     for specific functional groups not now
     addressed by the method.
     to evaluate use of the model to esti-
     mate chemical mixture permeation.
   This  report was submitted in fulfillment
of Contract  68-03-3293, Work Assignment
3-80, by Arthur D. Little, Inc.,  under spon-
sorship of the U.S. Environmental Protec-
tion Agency.

-------

-------