Permeation studies of PVC pipes with near infrared ... - SAGE Journals

L.E. Agelet et al., J. Near Infrared Spectrosc. 15, 283–289 (2007)�

�� 283

Permeation studies of PVC pipes with near infrared spectroscopy Lidia Esteve Agelet,a,* Charles R. Hurburgh,a Feng Mao,b James J. Gauntb and Say Kee Ongb Department of Agricultural and Biosystems Engineering, 1545 Food sciences building, Iowa State University, Ames, IA 50014, USA. E-mail: [email protected] a

b

Department of Civil, Construction and Environmental Engineering, Iowa State University Ames, Iowa, USA

Polyvinyl chloride (PVC) pipes are commonly used to transport drinking water. Although PVC is resistant to natural environmental conditions, organic solvents may attack the pipe wall causing swelling, softening, water lines failure and drinking water pollution. Leaks from underground storage tanks and random accidental spills of organic solvents or fuels place the pipes in contact with organic solvents. Near infrared (NIR) spectroscopy was used to track the permeation of PVC pipes by three major organic solvents (toluene, benzene and gasoline) at different concentrations. Partial least squares (PLS) calibrations with NIR spectra and reference data gave accurate models with R2 > 0.9, relative performance determinant (RPD) > 3 and low standard errors of prediction (SEP). These models could predict the permeation status measured by mm of solvent moving front, weight gain, or days under permeation. A second study correlated pipe permeation susceptibility to pure toluene in mm h–1/2 to the pipe spectra. Spectra differences from several pipe brands and sizes were modelled with locally weighted regression (LWR), resulting in models with accuracy (RPD) of around 5. NIR was a suitable tool to evaluate the permeation of PVC pipes and to predict the susceptibility of PVC pipes to permeation. Keywords: permeation, PVC, pipes, organic solvents, near infrared spectroscopy

Introduction Permeation of chemicals through plastic pipes and gaskets is a proven cause of drinking water contamination in the United States. Most cases (89%) have been associated with high levels of contamination in soil due to gasoline, oil or other products (including gasoline derivatives) and have occurred in areas of high risk for chemical spillage.1 In Montana, for instance, there have been more than 4000 known petroleum releases that polluted soils; some affected water pipelines, such as a case in 2002 where benzene permeated a polystyrene (PE) pipeline. 2 Texas had the highest releases of benzene and toluene, 1,135,994 and 16,285 pounds respectively, from 1987 to 1993.3 Rapid real estate development and a shortage of undisturbed sites favour building on already polluted soils where risk of pipe permeation may be higher. Since plastic pipes are easy to install, do not show corrosion and have a long service life, they are popular materials for water lines. Chemical permeation of PE and polyvinylchloride (PVC) accounted for 39% and 15%, respectively, of the total permeation cases reported in the United States.1 PVC seems less susceptible to permeation, although cases

DOI: 10.1255/jnirs.737

of permeation have been reported at high levels of solvent contamination in the soil.4 Once the material has been permeated, higher rates of chemical diffusion occur. Pipes become more susceptible to permeation if a second spill occurs, due to the chemical degradation and higher porosity of the material after pipe swelling.4 Few instruments are available to measure permeation variables. 5 Development of near infrared (NIR) spectroscopy prediction models for PVC permeation may become a new technique to get permeation parameters. A good prediction model would be an important tool for future research of permeation under different environmental conditions. Although permeation affects physical properties, changes in molecular bonds or vibrations due to the solute passing through the pipe wall is expected. Solute concentrations themselves will give different NIR spectra and thus are a way to detect the extent of permeation of the pipe. A discrimination analysis using NIR spectra may allow prediction of the future performance of pipes when exposed to organic solvents. Fast scanning and non-destructive analysis make NIR spectroscopy an attractive tool for permeation prediction and understanding.

ISSN 0967-0335

© IM Publications 2007

284 �� Permeation Studies of PVC Using NIR

• •

This study had two objectives: use NIR spectroscopy to track permeation with common solvents at varying concentrations develop an NIR method to predict susceptibility to permeation of new PVC pipes.

Materials and methods Tracking PVC pipes permeation study Samples Three 2.5 cm diameter PVC pipes, each from a different manufacturer, were used. The ends of the PVC samples (about 12 cm long) were sealed with square pieces of glass with waterproof epoxy sealant. Before the second piece of glass was sealed to the pipe, the pipe was filled with water to about 1.25 cm from the top. The length of pipe exposed to the solvent was about 10 cm and about 1 cm was covered with the sealant from both ends. The area of each pipe sample exposed to the solvent was the same for all the samples and experiments (Figure 1). Solvents and reference data Solvents were those involved in current permeation incidents: toluene, benzene and gasoline.1 Toluene and benzene (99 and 99.5% purity respectively) (Fisher Scientific International, Inc., Fair Lawn, NJ, USA) were tested as pure solvent, as well as 20, 40, 80 and 100% aqueous saturated solution. Gasoline was premium grade gasoline purchased at ConocoPhillips station (Ames, IA, USA) and only tested as pure solvent. Duplicate samples were run for each solvent at each concentration for a total of 66 samples. The PVC pipe samples were vertically immersed in individual glass jars filled with solvent at the designated concentration, ensuring no bubbles or free space in the jar with a Teflon cap fitted. The proportions of aqueous solution were achieved by weight using a Masterflex peristaltic

Figure 1. Two samples sealed with glass and epoxy.

tubing pump (Cole–Parmer Instrument Company, Vernon Hills, IL, USA) with Teflon tubes. The solutions were replaced every week to ensure constant concentration during the study. Two laboratory reference measurements of permeation were developed in the Iowa State University Environmental Engineering laboratory; moving front expressed as the moving front thickness (mm) and solvent sorption expressed as weight gain in grams cm–1 or % by weight.6 The moving front test measured the progress of the solvent front through the pipe to the nearest 0.001 mm, using an Olympus BHM microscope (Olympus America Inc., Center Valley, PA, USA). The microscope identified the swollen and darker region caused by the solvent flux through the pipe wall. The sorption test determined the weight gain per time when the pipe was exposed to solvent. Both sorption and moving front data were used as references for pure toluene and benzene calibrations. For aqueous solutions and for gasoline, permeation was slower and not measurable in the short term in the laboratory. Reference data for these tests were times under soaking, in days. The standard error of the laboratory (SEL) was calculated as the root mean square error from the analysis of volume (ANOVA) using SAS v 9.1.3 (SAS Institute Inc., Cary, NC, USA). SEL was the variability among replicate readings in the moving front test or jars in the sorption test, pooled across days—readings or days—jars for the moving front and sorption test, respectively. NIR equipment and spectra measurements A monochromator-based instrument (Foss NIRsystems 6500, Foss-NIRsystems, Laurel, MD, USA) was used for the study to scan from 400 nm to 2500 nm, in 2 nm increments. Reflection data as log (R–1) were taken. The instrument was using the natural products cell, programmed to scan only 25% of the travel distance to avoid the sealant and the tape at the ends. Samples in pure solvent were scanned every day (six scans per day) for seven days. Because permeation for solutions was lower, samples in aqueous solutions were scanned every two weeks for seven months, four times per sample for each interval. Samples soaked in gasoline were scanned every two weeks for nine months with six scans per sample. Only two sides of each pipe were scanned due to the sample conformation after sealing, manufacturer labelling (brand marks) and manipulation issues. Complete permeation was not achieved for either pure or aqueous solutions. Solvents did not reach the internal pipe surface. Spectral data were taken limited to wavelengths from 1102 –2500 nm. This range includes first and second CH overtone regions (around 1650–1790 nm and 1150–1250nm, respectively) and CH combination regions (around 2200– 2450 nm). The third overtone region (around 850 –950 nm) was excluded because it did not add useful information in

L.E. Agelet et al., J. Near Infrared Spectrosc. 15, 283–289 (2007)

285

preliminary tests. This also excluded variability between detectors in the instrument.

N is the number of reference samples. Three pipes were replicated.

Calibration models and spectra pretreatment Partial least squares (PLS) calibrations have been used in previous polymer and organic solvent studies.7–9 PLS models were developed for tracking permeation, using both raw data and pretreated spectral data with first and second Savisky–Golay derivatives (five points of smoothing and polynomial of third order). Validation of these models was done with cross validation: 120 to 604 scan groups, depending on the available data size in each experiment. PLS calibrations were developed with Unscrambler v 9.5 (Camo Technologies, Trondheim, Norway). Model accuracy was reported as: coefficient of determination (R 2) of the calibration, standard error of the crossvalidation (SECV) and relative performance determinant (RPD) from the validation.

NIR equipment and spectral measurements A Foss NIRsystems 6500 in reflectance mode was also used for predicting pipe permeation susceptibility, using the same spectral range as the permeation tracking study (1102 –2500 nm). The original samples were scanned six times each, except that samples larger than 3.2 cm in diameter could only be scanned twice, since they had to be cut to fit the natural cell device. The total number of scans from the two sets, with triplicates, was 828.

Predicting permeation susceptibility of PVC pipes Samples Samples for the prediction were purchased in two sets of 28 and 30 pipes, respectively, according to their applicability for drinking water main waterlines, manufacturer, diameter and production date. The two sets included seven diameters and five manufacturers. With three 12 cm triplicate samples for each type of pipe, there were 84 samples in the first set and 90 samples in the second set. Solvents and reference data The reference data for permeation susceptibility were obtained from 48 hour moving front tests with pure toluene (referred to as fast moving front test). Samples from each pipe were soaked in pure toluene for up to 48 hours. The toluene moving front was read at six hour increments. For each pipe, the readings of the moving front test were linearly related to the square root of time under soaking with R2 above 98%.10 The slope of the line (mm h–1/2) was characteristic of each pipe; this slope was taken as the permeation susceptibility reference parameter. The fast moving front test was developed using pure toluene only, since pure toluene permeation was fast yet toluene is less dangerous than benzene. Pipe preparation for this test was the same as in the permeation rate study. SEL for the permeation susceptibility parameter was calculated applying the SEL equation suggested by Marten et al.:11

where: y_ij is the ith replicate and jth sample Y ij is the mean of the replicates for the jth sample R is the number of replicates per reference sample. Five replicates were used.

Calibration models and spectral pretreatment PLS models were used for the susceptibility study, with raw and pretreated spectral data. The set of 28 pipes was used for calibration and the second set of 30 pipes was used for validation. These models were grouped as M1. The second group of PLS models, referred to as M2, were developed using 75% of the spectral data from the two sets. The remaining 25% were used for validation. The same spectral data sets as used for M2 (75% of the spectral data for calibration and the remaining 25% for validation) were used to develop locally weighted regression �� (LWR) models �� with principal component regression (PCR). Standard normal variate (SNV) pretreatment was added to the LWR model choices. PLS calibrations were developed with Unscrambler v 9.5 (Camo Technologies, Trondheim, Norway). LWR-PCR used Matlab v 7.0.4 (Mathworks, Inc., Natick, MA, USA) and PLS_Toolbox v 3.5.4 (Eigenvector, Inc., Seattle, WA, USA). The LWR function requires a number of local points (samples) for the model. This number was iterated as 25, 50 and 75. The number of PCs was iterated from 5 to 15. Model accuracy was reported with reference to the coefficient of determination (R 2), SECV, SEP when an external set was used for validation and RPD from the validation.

Results and discussion Tracking permeation through PVC pipes Permeation of pure benzene and toluene with NIR spectroscopy over seven days could be tracked with NIR (Figure 2) and gave accurate models (Table 1). RPD values were high for either raw or pretreated spectral data. Raw data and first derivatives usually gave the best results. SEL values were low (less than 50% of SECV except for the toluene sorption test). Since samples tested in the laboratory required sample modification or destruction, independent samples were tested with NIR. This could have contributed to higher variability from external sources in this study and, thus, higher SECV values. A larger number of PCs were used because models needed to account for spectral differences among pipes as well as solvent concentrations in the pipe wall.


1.8 1.6 1.4

Log(1/R)

1.2 1 0.8 0.6

Spectra for pipe unexposed to toluene

0.4 0.2 800

1000

1200

1400

1600 1800 wavelengths, nm

2000

2200

2400

2600

Figure 2. Spectra evolution in one of the PVC pipes soaked in toluene for nine days.

Data from pure toluene and benzene were combined to form general models. The general models had less predictive ability, as expected, but RPD were still above 3—usable for rough screening.12 The same number of PCs was used in the general models. For all the pure toluene and benzene models, the regression coefficients with more relevance in the calibration were those located in the first overtone region for the CH aromatic bonds (1600–1700 nm), C–H and C–C combination region (2100–2200 nm), CH3 combination region (2250– 2350 nm) and variables located in CH–CH combination region (2440–2460 nm). This was expected because of the chemical structure of these solvents 13 and PVC, both of which have CH bonds. Since the models must also account for differences between pipe brands, other random peaks

may refer to individual characteristics of each pipe type. Pure benzene models showed peaks in the region around 1350–1450 nm (in the CH first overtone region) possibly from solvent absorption. Permeation was not detected by moving front or weight gain for pipes soaked in gasoline for more than one year. However, NIR was able to identify changes after the first two weeks. The calibration showed a very strong linear relationship with days of exposure, using raw spectral data (Table 1). Time of exposure had to be used for gasoline and solvent solutions because the laboratory methods were not showing changes, while the NIR spectra were. The spectral data from the four aqueous solutions (either toluene or benzene), seven months, three pipes with duplicates, were related with the time under permeation conditions. Time as an independent variable was used for all solvent concentrations. The best toluene model, including all pipes and concentrations gave a 19 PC model with R2 = 94.5%, SECV = 16.11 days and RPD = 4.12 (Table 2). Pretreatment with derivatives did not improve the results. Benzene models showed the same characteristics as toluene models. The best benzene model used first derivative pretreatment and gave a model with 16 PC, R 2 = 95.1%, SECV = 16.12 days and RPD = 3.99 (Table 2). Data were then divided by permeating solution for both toluene and benzene. Models were developed for each aqueous solution and solvents. Table 2 shows how water apparently added uncontrollable variability. Water may interfere in the analysis, as the high weight of regression coefficient corresponding to wavelengths around 1950 nm (water first overtone region) showed. SECV increased as solvent solution decreased. Lower concentrations showed possible non-linearities.

Table 1. Best NIR models for PVC pipes exposed to pure solvents.

Model

Reference data

Pretreatment

Gasoline

Days

Pure toluene

Pure benzene

Pure benzene and toluene

R2

RPD

SECV

PCs

SEL

Raw data

0.99

8.8

9.27

7

n.a.

Moving front (mm)

1st derivative

0.99

13.0

0.07

6

0.02

Sorption test (g cm–1 pipe)

1st derivative

0.99

11.4

0.12

7

0.11

Sorption test (%)

1st derivative

0.99

10.8

1.44

7

1.08

Moving front (mm)

Raw data

0.99

9.7

0.08

9

0.01

Sorption test (g cm–1 pipe)

Raw data

0.98

�� 6.9

0.17

9

0.07

Sorption test (%)

1st derivative

0.98

�� 6.7

2.13

6

0.24

Moving front (mm)

2 derivative

0.96

�� 4.7

0.18

9

0.02

Sorption test (g cm-1 pipe)

Raw data

0.96

�� 5.3

2.65

9

0.09

Sorption test (%)

Raw data

0.98

�� 8.1

2.22

9

0.68

nd


287

Table 2. Best models by sets to predict permeation performance of PVC pipes exposed to pure and diluted solvents

Solvent

Solution

Toluene

Benzene

Benzene and toluene

R2

RPD

SECV (days)

PCs

General model

0.94

4.2

16.11

19

100% saturated solution

0.97

6.0

10.79

�6


0.98

7.7

�� 8.74

�8


0.95

4.2

16.38

�9


0.94

3.8

17.76

11

General model

0.95

4.0

16.12

16


0.98

7.1

8.85

�8


0.96

4.8

13.44

�8


0.96

4.8

13.65

15


0.95

4.5

15.15

11

General model

0.91

3.2

20.21

19

Exposure times were not predicted as well as in the higher concentrations. Aqueous solution data from toluene and benzene were combined to develop a general model. The model with raw data gave a 19 PCs model with R2 = 90.8%, SECV = 20.21 days and RPD = 3.25 (Table 2). First derivative usage improved the results only slightly. The accuracy was not sufficient to use the combined model for purposes other than rough screening (prediction of approximate days under general solvent permeation conditions with low accuracy).

Prediction of future permeation performance of PVC pipes Permeation performance was estimated in three ways:

•

M1 models—global PLS for 28 pipes, 30 pipes for v alidation • M2 models—global PLS for all pipes combined. 25% of spectra used for validation • M3 models—locally weighted regression with principal component regression for all pipes combined. 25% of spectra used for validation. The independent variable was the rate of advance of the toluene-moving front (k) in mm h–1/2. Calibration models M1 showed high correlation of spectra to k for both raw and treated spectral data (Table 3). However, these models could not predict the second set with accuracy. M2 models used a common method for validation of NIR calibrations, sub-setting the entire population. M2 model sta-

Table 3. Models to predict the permeation performance of PVC pipes exposed to toluene

M1 calibration models (Set 1)

Validation (Set 2)

PCs

R2

RPD

SEP (mm h–1/2 )

Raw data

19

0.96

1.0

0.02

1 derivative

16

0.93

1.0

0.01

2 derivative

12

0.89

1.1

0.01

Data treatment

st

nd

M2 calibration models (75% of whole data) Data treatment

Validation (25% of whole data)

PCs

R2

RPD

SEP (mm h–1/2 )

Raw data

18

0.82

2.3

0.01

1st derivative

16

0.87

2.2

0.01

2 derivative

13

0.86

1.9

0.01

nd


tistics improved over M1 but �� were still �� not useful. The clustering effects detected in the principal component analysis (Figure 3) was due to pipe types and this seemed to be one of the reasons why these models failed and used a high number of PCs. The SEL value was 0.0018 mm h–1/2, so laboratory reproducibility was not limiting the accuracy of the models. LWR-PCR models (M3) were then iteratively developed. LWR-PCR will identify those calibration samples most similar to a new sample and then create a model specifically for that sample. There was not a single solution or best model among the iterations. Multiple combinations of PCs, neighbour samples and pre-treatments gave similar accuracies. The trend of these models was that, for the same accuracy, the more samples included in the LWR-PCR models the more PCs must be included in the calibration. Since more PCs can create over-fitting, models with good performance and few PCs were preferred. Pretreatments with derivatives gave poorer results, while SNV models showed better accuracy than raw data models. For SNV, when 25 neighbour samples were used, the best results were acquired with five to ten PCs (Figure 4); these had the highest values of RPD at six, seven and eight PCs. The best RPD values were above 7, which was much better than either the M1 or M2 models. NIR predicted PVC pipe susceptibility to permeation with accuracies suitable for quality control with LWR-PCR, where traditional global PLS models failed.

Conclusions NIR spectroscopy successfully tracked permeation by toluene and benzene in PVC pipes. All reference data [mm, weight gain (%) and weight gain (g cm–1)] showed low laboratory standard error and gave accurate and predictive partial least squares (PLS) models. When time (days of soaking) was used as the reference data, solvent saturated solutions

- 1/ 2

Surface of SEP values (mm h

) for SNV pretreatment

0.008 0.007 0.006 0.005 0.007-0.008

0.004

0.006-0.007

0.003

0.005-0.006

0.002

0.004-0.005

0.001

0.003-0.004

0 5

0.002-0.003 6

7

0.001-0.002 8

9 Pcs

0-0.001 10

75 objects

11

12

50 objects

13

14

15

25 objects

Figure 4. Surface of SEP values for SNV pre-treatment for M3 models from the susceptibility study.

gave models accurate enough for rough screening. Lower solvent concentrations lead to higher error of prediction. Permeation by gasoline could be modelled with PLS with good accuracy. NIRS could identify changes in the pipes soaked with gasoline as early as day 15, while traditional laboratory methods could not detect permeation after more than one year of soaking. NIR detected the high variability among original pipes and among pipe types. Spectral differences were correlated with permeation susceptibility as obtained from the fast moving front test. Although the heterogeneity and diversification of pipe population could not be modelled with global PLS, locally weighted regression models provided accurate models with maximum RPD values around 7. Because NIR spectroscopy acceptably predicted pipe permeation only with LWR, any new samples must belong to pipe types included in the calibration population. Database expansion would be needed for continued use.

2

Acknowledgements 1.5

PC 2 scores

1

0.5

0 -3

-2

-1

0

1

2

3

4

5

-0.5

Iowa State University gratefully acknowledges that the Awwa Research Foundation is the joint owner of the technical information upon which this manuscript is based. Iowa State University thanks the Foundation for its financial, technical and administrative assistance in funding and managing the project through which this information was discovered. The comments and views detailed herein may not necessarily reflect the views of the Awwa Research Foundation, its officers, directors, affiliates, or agents.

-1

References

-1.5

PC 1 scores

Figure 3. Clustering pattern from the PCA for all pipes from the susceptibility study.

1. T.M. Holsen, J.K. Park, D. Jenkins and R.E. Selleck, “Contamination of potable water by permeation of plastic pipe”, J. AWWA 83(8), 53 (1991).


2. Department of Environmental Quality of Montana, Department of Environmental Petroleum Release Section (DQ-PRS), Permeation of Waterlines by Petroleum Constituents. Technical Guidance Document #16. Retrieved July 2006 from DEQ web site: http://deq. mt.gov/LUST/TechGuidDocs/Techguid16.pdf 3. Environmental protection agency, “Consumer Factsheet on: TOLUENE”. Retrieved May 02, 2006, from EPA web site: http://www.epa.gov/safewater/contaminants/ dw_contamfs/toluene.html 4. American Water Works Association (AWWA), Permeation and Leaching. Document prepared by AWWA with assistance from Economic and Engineering Services, Inc. (2002). Retrieved August 2006 from EPA web site: http:www.epa.gov/safewater/tcr/pdf/permleach.pdf 5. J. Comyn, “Polymer permeability”, J. Polym. Sci., Part C: Polym. Lett, 24(8), 431 (1985). doi: 10.1002/ pola.1995.080331701 6. J. Gaunt, F. Mao and S.K. Ong, Performance of plastic and pipe gaskets in hydrocarbon contamination: Field experience and laboratory studies. Paper presented at the 2006 meeting of AWWA Annual Conference Proceedings (CD), Session Thu7, paper 6. American Water Works Association, Denver, CO, USA (2006). 7. D. Fischer and K.-J. Eichhorn, “Process analysis of polymers by NIR spectroscopy”, Analusis 26(4), M58 (1998). 8. D. Garcia, F. Borden and E. Bouveresse, “Identification, quantitation and qualification of PVC multifunctional systems with near-Infrared spectroscopy and

289

9.

10.

11.

12.

13.

chemometrics”, J. Vinyl Additive Tech. 7(4), 214 (2001). doi: 10.1002/vnl.10293 S.E. Barnes, E.C. Brown, M.G. Sibley, H.G.M. Edwards and P.D. Coates, “Vibrational spectroscopic and ultrasound analysis for the in-process monitoring of poly(ethylene vinyl acetate) copolymer composition during melt extrusion”, The Analyst 130, 286 (2005). doi: 10.1039/b416244g F. Mao, Permeation of organic contaminants through polyvinyl chloride (PVC) and polyethylene (PE) pipes. Unpublished Phd thesis, Iowa State University, Ames, IA. USA (2007). G.C. Marten, J.S. Shenk and F.E. Barton. Near Infrared Reflectance Spectroscopy (NIRS): Analysis of Forage Quality. Agricultural Handbook 643, USDA-ARS, Washington DC, USA (1989). P. Williams and K. Norris. Near-Infrared technology in the agricultural and food industries, Ed by P. Williams. and K. Norris. American Association of Cereal Chemists Inc., St Paul, Minnisota, USA, pp.165 (2001). L.G. Weyer and S.-C. Lo, “Spectra-structure correlations in the Near-Infrared”, In Handbook of vibrational spectroscopy, Ed by John M. Chalmers and Peter R. Griffiths. John Wiley & Sons Ltd, Chichester, UK, pp. 1817 (2002). Received: 7 March Revised: 11 April 2007 Accepted: 13 April 2007 Publication: 18 September 2007