EPA/600/A-92/201
FLEX: AN EXPERT SYSTEM FOR REVIEWING THE CHEMICAL RESISTANCE
OF FLEXISLE MEMBRANE LINERS FOR HASTE MANAGEMENT
Robert E. Landreth
Chief, Landfill Technology Section
Municipal Solid Waste and Residuals Management Branch
Waste Minimization, Destruction and Disposal Research Division
Risk Reduction Engineering Laboratory
United States Environmental Protection Agency
Ci ncinnati , Ohio 45268
ABSTRACT
A computerized, expert system (Flexible Uner Evaluation eXpert, FLEX) has been developed that will
assess the chemical resistance of flexible membrane liners (FML) used at waste management facilities.
These FML's must be chemically resistant to the waste, its leachates, or both. To demonstrate chemical
resistancy, the United States Environmental Protection Agency (USEPA) requires the use of Method 9090,
wherein the physical properties of the FML are evaluated after various exposure times.
After the Method 9090 data are assembled, a reviewer is then responsible for assessing the data and
reaching a conclusion about the probability that a proposed FML material will be resistant to the waste/
leachate. Knowledgeable judgments as to the chemical resistance of FML require a multitude of consider-
ations—considerations whose interrelationships can only be understood, if at all, by persons having a
combination of theoretical and practical experience in chemistry, polymer science, waste site construction,
etc. Because such knowledge is not generally available within the USEPA, a computerized, expert system
was sought to provide guidance in the review of Method 9090 data.
Several FML experts were interviewed to capture the thought processes they used when assessing the
chemical resistance of liner materials. The liner materials of interest were high-density polyethylene
(HOPE), polyvinyl chloride (PVC), and chorosulphonated polyethylene (CSPE). The results of these inter-
views were reduced to simple criteria or "rules' for chemical resistance that could be expressed in the
form of a computer program. The resulting FLEX program has been Beta tested and is now undergoing final
review before release to the public. This paper describes the development of the expert system, the BETA
testing to verify the system, and experiences in using the system.
The system is programmed to run on an IBM PC-XT, AT or 1001 compatible, that has a minimum of 512K
bytes of RAM, DOS version 2.0 or higher, a printer configured as system device C: with at least 1.5 mega-
bytes of free space.
-------
TECHNICAL REPORT DATA
(Please read Instructions on the reverse before compter
1. REPORT NO. 2.
EPA/600/A-92/201
3
«. TITLE AND SUBTITLE
FLEX: An Expert System for Reviewing the Chemical
Resistance of Flexible Membrane Liners for Waste
Management
5. REPORT DATE
6. PERFORMING ORGANIZATION CODE
7. AUTHOR(S)
Robert E. Landreth, (513)569-7871
B. PERFORMING ORGANIZATION REPORT NO.
9. PERFORMING ORGANIZATION NAME AND ADDRESS
USEPA/RREL/WMDDRD/MSWRMB
5995 Center Hill Avenue
Cincinnati, OH 45224
10. PROGRAM ELEMENT NO.
11. CONTRACT/GRANT NO,
12. SPONSORING AGENCY NAME AND ADDRESS
Risk Reduction Engineering Laboratory—Cincinnati, OH
Office of Research and Development
U.S. Environmental Protection Agency
Ci nci nnati, OH 45268
13. TYPE OF REPORT AND PERIOD COVEREO
Published Paper
14. SPONSORING AGENCY CODE
EPA/600/14.
15. SUPPLEMENTARY NOTES
Project Officer = Robert E. Landreth (513) 559-7871
2nd International Symposium on Environmental Geotechnology, Tongui University, Shanghai,
China
16. ABSTRACT
5 A computerized, expert system (FLEX) has been developed that will assess the chemical
resistance of flexible membrane liners (FML) used at waste management facilities.
These FML1s must be chemically resistant to the waste, its leachates, or both. To
demonstrate chemical resistancy, the United States Environmental Protection Agency
(USEPA) requires the use of Method 9090, wherein the physical properties of the FML
are evaluated after various exposure times.
After the Method 9090 data are assembled, a reviewer is then responsible for assessing
the data and reaching a conclusion about the probability that a proposed FML material
will be resistant to the waste/leachate. Knowledgeable judgments as to the chemical
resistance of FML require a multitude of con si derations,--considerations whose
interrelationships can on1_y. b,e, u.nde-r-s-t-eod-r=i f--at=a-l Vv"'"by persons having a combination
of theoretical and practical experience in chemistry, polymer science, waste site
construction, etc. Because such knowledge is not generally available within the USEPA,
a computerized, expert system-was sought to provide guidance in the review of Method
9090 data.
Several FML experts were interviewed to capture the thought processes they used when
assessing the chemical resistance of liner materials. The liner materials of interest
were high-density polyethylene (HOPE), polyvinyl chloride {PVC), and c-horosul phona.tedv
polyethylene- (CSPE).^ — -
17. KEY WORDS AND DOCUMENT ANALYSIS
a. DESCRIPTORS^
b.IDENTIFIERS/OPEN ENOED TERMS
c. COSATI Field/Gioup
Chemical Resistance
Expert System
Geosynthetic
Landfills
<
18. DISTRIBUTION STATEMENT
RELEASE TO PUBLIC
19. SECURITY CLASS (This Rtportj
UNCLASSIFIED
21, NO. Of PAGES
12
20, SECURITY CLASS {This page)
UNCLASSIFIED
22. PRICE
EPA Fwm 2220-1 (R»*. 4-77) previous edition h obsolete
-------
INTRODUCTION
When a landfill or lagoon site is proposed, the U.S. Environmental Protection Agency (EPA) requires
the party seeking the permit to submit evidence that the proposed site will not cause damage to the sur-
rounding environment. To protect the soil and groundwater beneath and adjacent to the site, flexible
membrane liners (FML) have been determined to meet the Agency's liquids management strategy. Since there
are wide variations in the quality of leachates from waste management sites and a wide variety of candidate
liners, part of the proposal package must include data that will allow the EPA or state permit reviewer to
judge the suitability of the proposed FML material. Will the FML be chemically resistant to the antici-
pated wastes and leachates at that particular site?
To demonstrate the chemical resistance of the FML, EPA requires that its Method 9090 be used, as a
minimum, to test chemical resistance. In the test, the physical properties of FML specimens that have
been immersed in waste/1eachate (at 25°C and SO0C) are measured after each of four successive months and
compared with the original unexposed FML physical properties. These results are included in the applica-
tion for a permit to install and operate the site, and the EPA or state permit reviewer employs them to
assess whether the proposed FML material will be chemically resistant.
The decision of FML suitability is complex, requiring knowledge of chemistry, polymer science, waste
site construction, etc. Because a less complex decision-making process was needed for chemical compati-
bility determinations, EPA developed a computerized expert system--one that would embody the best thinking
of recognized experts in this field. Three materials were selected for inclusion in the system: polyvinyl
chloride (PVC) , chlorosulphonated polyethylene (CSPE), and high density polyethylene (HDPE).
To that end, FML experts at liner fabrication firms, liner material suppliers, and an independent
consultant were interviewed. What measures would these experts use and what was their rationale to evalu-
ate the chemical resistance of the proposed liner material? Their rationale became the criteria or rule
base for chemical resistance measurements that were expressed in the form of IF—THEN statements of a
computer program on an IBM-PC-AT employing sophisticated system shell based on the science of artificial
i ntel1i gence.
The result is the Flexible Uners Expert system: The FLEX system for evaluating chemical resistance
data of FML's. Unlike conventional computer software which produces numeric results, the results of an
analysis by the FLEX system are presented in a written textual report generated as the system runs. This
report lists problems with the data (too scattered or missing) as well as any values that indicate that
the liner may be substandard or incompatible with the immersion medium. Also included in this report are
explanations as to why any data are deficient. The report may therefore be useful to a Method 9090 test
reviewer to confirm a decision IF the review concurs with the rationale proposed by the FLEX system.
The system is programmed to run on an IBM PC-XT, AT or compatible, that has a minimum of 512 bytes of
RAM, DOS Version 2.0 or higher, a printer configured as System Device C: with at least 1.5 megabytes of
free space.
2
-------
It is important for the FLEX system to be able to discern whether or not Method 9090 data input by
the user is reasonable before applying sore specific rules. Originally, the system used a two part method.
The first part checked to see if the standard deviation of the samples in any time period was greater than
101 of the mean of these samples. If this 10* threshold was breached, then the data was considered scat-
tered and no further analyses were performed. If scatter was not detected, a quadratic regression was
performed on the data to determine the samples of each time period. It was conceivable that the data
could be so scattered that it would be meaningless but the means conform nicely to a quadratic curve. In
this case, the regression analysis would indicated a problem, thus some scatter determination is necessary
before the regression.
The 1 OS threshold was an arbitrary decision. A more scientific approach was implemented using a
test for lack of fit in the current version of FLEX. The lack of fit test compares the variance in the
samples for a time period to the variances in the samples across time periods (control through month four).
If the variance in a time period is more significant than the variances across time, then the data are
considered scattered. The sensitivity of this test is adjustable. The problem with this test is that if
the test results are relatively consistent throughout time (which one would expect for a chemically resis-
tant liner) then a small deviation within a time period will trigger a scatter determination.
The statistics that are currently used in the FLEX program are at the end of this paper.
It should be noted that the program is designed to PROVIDE ASSISTANCE to those responsible for evalu-
ation of Method 9090 test results. Under no circumstances should the recommendations of FLEX be considered
absolute; especially when the FLEX system finds no problems with the test results. FlEX is not a substi-
tute for a review of the test results by a skilled professional. Rather, FLEX is a screening tool to be
used by those familiar with flexible membrane liner testing and EPA Method 9090. The system can rapidly
pinpoint inconsistencies in the test data and test results that suggest the liner is substandard or incom-
patible. In this capacity, FlEX can save time, reduce oversights, and enhance the consistency of Method
9090 test reviews.
THE COMPARISON TESTING
To test the validity of the FLEX advisory system, five recognized FHL experts evaluated sample data
sets for three types of FMl's--data sets similar to those produced by EPA Method 9090 (Table 1).
The 14 different sample sets of the three types of liner materials included data measurements of
properties used to judge the chemical resistance of FHL liners: unexposed specimens and exposed specimens
measured at 25°C and 50"C for each of 4 months. Common to all 14 samples were percent-change measurements
for weight, 1001 modulus, and elongation at break. Each sample set did not include data for all the prop-
erties (Table 2). This lack of submitted data was typical before Agency policy established which data
were required.
At least two experts analyzed each type of "Nl (Table 3).
3
-------
Table 1. Immersion-test data sets for three
flexible membrane liners*
No. Temperature
Abbre- of data of samples
Sample FML viation sets reviewed
High-density polyethylene HDPE 6 25°C 4 50°C
Polyvinyl chloride PVC 4 25°C 4 50°C
Chlorosulphonated polyethylene CSPE 4 25°C 4 50°C
~The data were not those of an actual Method 9090 test.
+The data for both 25°C and 50°C were submitted for each of the three FML's.
4
-------
Table 2. Properties used to judge chemical resistance
of three FNL's; all properties not tested
for each FML.
Property
Unit of
measurement,
J change
HOPE
PVC CSPE
Weight
Tensile strength at yield
Elongation at yield
1001 modulus
Breaking strength
Elongation at break
Length
Width
Puncture strength
Tensile strength at break
Tear strength
Hardness
Thickness
gram
psi
•
M
psi
psi
i nch
inch
pound
psi
psi
B-2
inch
All HOPE FHL samples were tested for this property,
+A11 PVC FML samples were tested for this property.
*A11 CSPE FML samples were tested for this property,
xTests conducted.
"Tests not conducted.
5
-------
I
Table 3. Expert and FML analyzed
Expert HOPE PVC CSPE
Fl£X x xx
A xxx
e x
c x
0 x
E - - x
"Sample analyzed,
"Sample not analyzed.
6
-------
The experts were asked to review each assigned data set and conclude that
° the results indicate that the liner materia! is not resistant to the chemical/waste {not
resistant), or
° there are no results that indicate the liner material is not resistant to the chemical/waste* (no
indication not resistant), or
° data are insufficient to reach a conclusion (data missing), or
° data are inconsistent (scattered).
For each conclusion, the expert was asked to give the reason for the conclusion, state which test
results were the most or least important in reaching the conclusion, draw trend lines or graphical exhibits
of data provided with the data, and state how the data at 25°C and 50°C were used. In addition, if the
conclusion was that there were insufficient data, the expert was to describe what additional data would be
needed to reach a conclusion, and if there were inconsistencies in the data, the expert was to identify
and describe how the inconsistencies could be worked around to reach a conclusion,
RESULTS
To judge the validity and limitations of the FLEX systen and to identify where it can be improved,
the conclusions and reasonings of the experts were compared with those of the system. In Tables 4, S, and
6, the areas of agreement and disagreement between the FLEX system and the FML experts are noted for the
three types of FML's.
There was good agreement between FLEX and Expert A for all three liner materials; there was less
agreement between FLEX and the other experts. Generally FLEX and Expert A were more apt to find the FML
"not resistant" than were the other experts, although the reasoning behind FLEX and Expert A's conclusions
sometimes differed. Experts B through E applied less stringent criteria to the data, were more willing to
judge inconclusive data in favor of "no indication not resistant,* and, in some cases, judged test results
considered important by FLEX and Expert A as outliers. The experts generally agreed about which proper-
ties were the most significant in judging chemical resistance: weight change and 1005 modulus (Table 7).
CONCLUSIONS
The comparison of the FLEX conclusions with those of the experts demonstrates the feasibility and
practicality of FLEX. FLEX is able to advise concerning the FML chemical resistance of data submitted as
part of a permit application.
The system can be characterized as conservative--more stringent in its criteria than were some of the
experts. Where FLEX and the experts disagreed, the system more often found the FML not resistant whereas
the experts judged the data as being missing or with no indication not resistant. The system is presently
useful to guide the permit reviewer and to identify those parts of the permit application where a more in-
depth investigation is needed.
•This statement does not say that the liner material is resistant.
7
-------
Table 4. Agreement/disagreement among FLEX and FMl
experts concerning HOPE (at 25°C)
Scattered/
No indica-
Not
Data
confl icti ng
tion not
Liner
resistant
mi ssing
data
resistant
SO mil
HK.8
FLEX-A-B*
-
-
-
HI.8
FLEX-A
-
-
3
HJ.8
-
FLEX-A
FLEX
B
HK.8
FLEX-A
B
-
-
60 nil
HL.6
FLEX-A
-
-
8
HH.6
-
FLEX-A
FLEX
FLEX-3
•FLEX is
the computer system;
A and 8 are
two experts.
Table 5. Agreement/diagreement among FLEX and FMl
experts concerning PYC (at ZS'C)
L iner
Not
resistant
Data
missing
Scattered/
confl icting
data
No indica-
tion not
resistant
PP.3
FLEX-A*
-
-
C
PQ.3
A
A
-
FLEX-A-C
PR.3
FLEX-A
A
A
C
PS.3
FLEX-A
-
-
r
•FLEX is the computer system; A and C are two experts.
8
-------
Table 6. Agreement/disagreement among FLEX and FML experts
concerning CSPE (at 25°C and 50°C)
Liner
^ ©
I
Not
resistant
Data
mi ss1ng
Scattered/
confl i cti ng
data
No indica-
tion not
resi stant
CC.3
25°
-
-
-
FLEX-A-0-E
CC.3
1/1
O
o
FLEX-A-E
-
-
0
CO.3
25°
FLEX-A
-
-
D-E
CD.3
50°
FLEX-A
-
-
D-E
CE.3
25°
FLEX-A-D
-
-
E
CE.3
50°
FLEX-A-D
-
-
E
CF.3
25°
FLEX-A-D
-
-
E
CF.3
o
O
u">
FLEX-A-0-E
-
-
-
•FLEX is the computer system; A, D, and E are three experts.
Table 7. Relative importance* of properties (test results)
to the experts judging the chemical resistance of
FML materials
Property
HDPE
PVC
CSPE
A
6
A
c
A E
Wei ght
1
1
1
2
1 1
100S modulus
2
2
2
5
3 2
Tensile strength at
break
1 east
4
-
3
least
Elongation at break
least
-
-
1
least
Tensile strength at
yield
3
3
-
-
-
Elongation at yield
4
least
3
-
-
Puncture
5
-
-
6
-
Tear
-
-
3
4
-
Length
-
-
-
7
-
*1 is most important; 7 is least important.
-------
WHAT NEXT?
Information gaps do exist; the dependence of the results on time, data trend analysis, and the rela-
tive importance of data stabilization need investigation. The need to test FMl at two temperatures should
also be investigated.
To further refine FLEX, the system should be reevaluated with the use of more data sets and with
insights gained from the experts' analyses. The precision and accuracy of Method 9090 data should be
examined. Should there be different criteria for different liner materials? for judging when physical
properties have stabilized? Has the composition of the waste/leachate in the test cell remained constant
during the exposure period? What are the consequences of the loss of volatile solvents? Shouldn't the
fomat of Method 9090 be the same as that of the FLEX system, with similar units used to express similar
¦•esults? These questions should be resolved.
Part of the answer to these questions can be addressed in a good laboratory quality control-quality
assurance program. The development of a QA/QC program will at least identify shortcomings at the laboratory
level. When data are submitted frcn an individual laboratory, at least the Agency will have some understand-
ing of the data quality.
AVAILABILITY
The Agency is currently reviewing the FLEX Program to determine if it is consistent with policy. The
Office of Research and Development is also working with selected Regional Offices for final Seta testing and
to determine the utility of the system. It is intended to make the program generally available to the EPA
Regional Offices and authorized states.
-------
TESTS OF S 1GN IF ICANCf FOR A OUftDRAT 1C MODEL
ftotat ion
J
3 » b0 + bj* ~ b^x*
B * [ b0 b, b2 )'
mj = I replicates at sample location i
f * i distinct sample locations (time periods)
n = total number of observations
Y • t *i yz ••• yn y
ft
* v.?,*/"
t - { y y . ,. y ]'
S
y, = ' ¦ i ,f
J • [ >i •••#„]'
x . l t, t,2
B - (X *X )"1 (K 'Y |
test for tack of F it
SSlof = (J-XB)'(J*X8)
SSpe - (Y-J)'(Y-J)
f • 5Sv/
------- |