Click for next page ( 40


The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 39
39 APPENDIX B Evaluation of Mechanical and Chemical Test Methods

OCR for page 39
40 CONTENTS Introduction ................................................................................................................................................................................... 41 Overview of Test Program ..................................................................................................................................................... 41 Strand Samples ....................................................................................................................................................................... 42 Mechanical Test Methods and Results .......................................................................................................................................... 44 Transfer Length Testing......................................................................................................................................................... 44 Pull-Out Testing..................................................................................................................................................................... 55 Surface and Chemical Test Methods and Results ......................................................................................................................... 69 Contact Angle Measurement................................................................................................................................................. 70 Examination under Ultraviolet Light.................................................................................................................................... 73 Testing pH .............................................................................................................................................................................. 73 Loss on Ignition...................................................................................................................................................................... 77 Loss in Hot Alkali Bath .......................................................................................................................................................... 78 Change in Corrosion Potential.............................................................................................................................................. 79 Surface Roughness ................................................................................................................................................................. 79 Corrosion Rate ....................................................................................................................................................................... 80 Organic Residue Extraction................................................................................................................................................... 81 Atomic Absorption and Colorimetric Analyses ................................................................................................................... 83 Evaluation of Test Methods ........................................................................................................................................................... 84 Variability in Test Methods ................................................................................................................................................... 84 Evaluation of Mechanical Test Methods............................................................................................................................... 84 Evaluation of Surface and Chemical Test Methods.............................................................................................................. 86 Summary of Test Method Correlation................................................................................................................................ 103 Precision Testing .................................................................................................................................................................. 108 Development of Thresholds......................................................................................................................................................... 110 Regression and Prediction Intervals.................................................................................................................................... 111 Thresholds Based on Regression with Single Predictor ..................................................................................................... 117 Thresholds Based on Regression with Multiple Predictors................................................................................................ 119 Summary....................................................................................................................................................................................... 121 References ..................................................................................................................................................................................... 122

OCR for page 39
41 Introduction identify those methods where some degree of correlation was indicated. For each source of strand, bond performance was A number of test methods were proposed for use as part of measured in terms of pull-out stresses, transfer lengths, or a quality control (QC) program to evaluate bond of pre- both. Accordingly, for the purpose of this analysis, the bond stressed concrete strand. These were classified as performance- performance was treated as the independent variable. The best based (i.e., mechanical) tests and surface and chemical tests. experimental design for estimating a correlation is to place The mechanical methods included: the design points as far apart as possible in terms of the inde- pendent variable. Thus, the optimal statistical design is to run Pull out from concrete, each test on strands that show a range of bonding performance. Pull out from Portland cement mortar, For the screening experiments high, medium, and low bond- Pull out from gypsum plaster-based mortar. ing sources were desired. However, efforts to obtain a very low bonding strand were not successful. Although reports of low- The surface and chemical methods included: bonding-strand incidents continue to surface in the precast concrete industry, "unused" samples of such strand remained Contact angle, elusive. Therefore, the screening tests on new strands were Examination under UV light, run on what are essentially high bond and intermediate bond pH, strands. Some strand from a project in India with apparently Loss on ignition, very poor bond properties was tested. However, the surface Loss in alkali bath, chemicals on this strand did not appear to be the same as for Change in corrosion potential, North American manufactured strand, and therefore, the test Corrosion rate, results could not be directly compared to the rest of the test Surface roughness, program. Organic residue extraction/Fourier transform infrared (FTIR) spectroscopical analysis, Elemental analysis, Correlation Testing AA, The second round of experiments was performed for confir- Visible light spectroscopy. mation and calibration purposes. This round involved running additional tests using those methods that showed promise in These tests, as well as transfer length tests, have been per- the screening experiments. These selected tests were conducted formed as part of this research program, which is reviewed in on five new strand samples. This complete data set was then detail in this appendix. The purpose of this program was to used to assess the correlation between the QC tests and bond determine if any of these proposed tests are applicable for use performance, and to determine if the tests were able to accu- in a QC program. The QC tests have been divided into two rately identify good and bad strand. It was also used as a categories, depending on the complexity and time required to basis for discussing pass/fail criteria for acceptable bond conduct the tests: Level I and Level II QC tests. A summary of performance. the tests methods, their QC Level, and the test objectives are given in Tables 2 and 3 of the Project Report. Precision Testing Overview of Test Program A third round of testing was conducted to determine the precision, that is, repeatability, of those methods showing To evaluate these test methods, several rounds of exper- good correlation with bond strength. This was intended to imentation were conducted: screening, correlation, and form the basis of precision statements to be included in pub- precision. lished test methods. Screening Testing Basis for Evaluation--Transfer Length and Pull-Out Tests The first round of experiments consisted of "screening" experiments. The objective for the screening experimentation Transfer length is the most reliable and realistic measure of was to eliminate those tests that are not helpful for predicting bond performance. During the screening testing, the evalua- bond performance. Thus, the first step in this round of test- tion of correlations between the pull-out testing and the bond ing was to estimate the correlation between each surface or performance were based on performance as measured with chemical test and bond performance; the second step was to transfer length tests conducted on the same source of strands.

OCR for page 39
42 Pull-out testing was conducted as part of the screening The strand sources included in testing for this program studies using three materials as the test matrix: a concrete, are listed in Table B-1. This table also includes the first slip a Portland cement mortar, and a gypsum plaster mortar. or 0.1-in. slip stress measured in concrete pull-out tests or in Based on comparisons with transfer length tests conducted mortar pull-out tests, depending on what was available. Each in this study and described in this document, the concrete pull-out stress is the average of the pull-out stresses from at pull-out test showed the best correlation with bond quality. least six individual pieces of strand. The bond stresses are cal- The surface and chemical test methods were evaluated in the culated based on the actual surface area and the embedment Screening Round based on the results of pull-out tests from length of the strand. concrete, again on strand samples from the same source. These strand sources fall into three groupings: historic, re- However, the evaluation of correlation of test results to bond cently manufactured, and OSU (Oklahoma State University). in the Correlation Round of testing was based on results from a mortar pull-out test program associated with NCHRP Historic Strand--This study initially identified samples of Project 12-60 Transfer, Development, and Splice Length for strand for testing from prior tests conducted at Kansas State Strand/Reinforcement in High-Strength Concrete. The Principal University (KSU) by Bob Peterman and at Stresscon by Don Investigator from this project supplied the strand samples Logan that cover a wide range of pull-out behavior. These are for this portion of the study. No concrete pull-out testing was referred to as "historic" strand and were manufactured between conducted in the Correlation Round of the experimental 1997 and 2004. Figure B-1 is a plot of first slip stress or stress program. at 0.1-in. end slip versus maximum stress from the data avail- able from historic concrete pull-out tests. When suggested minimum pull-out loads for acceptable Strand Samples bonding performance (suggested by Don Logan of Stresscon, based on a limited number of flexural beam tests conducted To assess the effectiveness of the mechanical and surface in the mid-1990s and his engineering judgment) are con- chemistry-based testing procedures, it was essential that verted to bond stresses, they are 425 and 955 psi for the first samples representing the range of possible performance be slip and maximum stresses, respectively. These thresholds evaluated. Since neither precasters nor strand suppliers were have been reproduced in Figure B-1. enthusiastic about associating themselves with poor-bonding strand, obtaining samples of strand from the lower end of the Recently manufactured strand--Figure B-1 also shows performance spectrum was difficult. the concrete pull-out performance of recently manufactured Table B-1. Strand sources. Strand Geometry Mortar Pull-Out Testing Concrete Pull-Out Testing (LBPT) Measured Lay Strand Diameter Pitch (Handed- 0.1-in. Slip 0.1-in. Slip Source ID Size (in.) (in.) (in.) ness) Location Date Stress (psi) Location Date Stress (psi) Historic Strand KSU-F 1/2 Special 0.524 7 5/8 Left -- -- -- KSU Mar 2004 241 KSU-H 1/2 Special 0.523 7 1/2 Left -- -- -- KSU Mar 2004 209 SC-F 1/2 0.503 8 Left -- -- -- SC May 1997 223 SC-H 1/2 Special 0.530 7 1/4 Left -- -- -- SC Nov 2002 472 SC-IS 1/2 0.501 7 Left -- -- -- SC Mar 2003 682 101 6/10 0.601 8 1/2 Left -- -- -- SC Oct 2004 241 Recently Manufactured Strand 102 1/2 0.501 7 1/2 Left KSU Jun 2005 315 KSU Jun 2005 441 103 1/2 0.503 8 Left KSU Jun 2005 397 KSU Jun 2005 944 151 1/2 Special 0.517 7 1/2 Left KSU Jun 2005 273 KSU Jun 2005 541 153 6/10 0.588 9 Right -- -- -- KSU/SC Jun/Aug 2006 142/406 OSU Strand 349 1/2 0.505 8 3/4 Left OSU Jun 2004 156 -- -- -- 548 1/2 0.500 7 5/8 Left OSU Jan-Feb 2004 623 -- -- -- 697 1/2 0.503 7 1/4 Left OSU May 2004 606 -- -- -- 717 1/2 0.500 8 Left OSU Feb 2004 206 -- -- -- 478 * 1/2 0.499 7 5/8 Left OSU May-June 2004 409 -- -- -- 960 * 1/2 0.500 7 1/2 Left OSU May-June 2004 409 -- -- -- * Samples designated 478 and 960 were from same source.

OCR for page 39
43 U -11/02 103-B-0.5 1200 D IT-11/02 TW-5/96 E A S3 S5 IS-3/03 J B G S4 C 102-A-0.5 S1 S2 103-B-0.5 1000 151-Z-0.5 F 102-A-0.5 H-11/02 Max Stress (psi) H NC-05/01 101-A-0.6 800 hollowcore F-5/97 600 Historical results at first slip NCHRP tests at 0.1-in. slip Max Threshold D-5/96 1st Slip Threshold 400 200 200 300 400 500 600 700 800 900 1000 Bond Stress at First Slip or 0.1-in. Figure B-1. Correlation between maximum stress and first observed or 0.1-in. slip stress for historic and recently manufactured strand. samples identified during this project. These recently manu- ning Portland cement mortar pull-out tests. Source 151 is the factured samples were obtained in large quantities for the same strand manufacturer as 102, but from a precaster in the purpose of this research and were used in the screening exper- eastern United States who had not reported any strand slip iments. Notice that the greatest variation in the recently problems. As received, the spool of strand from the east coast manufactured strand is not in terms of maximum stress but precaster had small areas of rust on the outer windings of the in terms of the early bond stress. spool. These outer windings were removed, and the tested The recently manufactured strand sources (102, 103, strand was taken from an inner winding on the same spool. and 151) were selected because initial testing indicated that The 102 and 103 strands are both 0.5-in. diameter. The 151 they represented a range of first slip pull-out performance. strand is 0.52-in. diameter. None of these strands had significantly low maximum load One additional recently manufactured strand was received pull-out performance. after the completion of the Screening Round of evaluation. The bond stress at 0.1-in. slip of Source 102, measured in This strand was supplied from India and was numbered 153. concrete pull-out testing performed as part of this project, is This strand was tested in portions of the correlation program, slightly above Logan's 425 psi threshold. Of the 31 sources pre- but was not included in the correlation analysis, since it was sented on this plot, 13 are to the left of Source 102. Only one of learned that the manufacturing processes used in its production these (D-5/96) was available in enough quantity to enable were markedly different from those used to produce strand in additional testing, and the condition of this strand is variable. the United States. As stated, the three recently manufactured sources of strand obtained for testing were identified as 102, 103, and 151. OSU strand--The sample sources used for the Correlation Source 103 is the strand used by Stresscon in their ordinary Round of testing were selected by Bruce Russell of OSU. production of precast/prestressed concrete. It is a source with These sources of strand had been tested in work performed a proven record of good bond from pull-out tests, flexure by Russell for the NCHRP Project No. 12-60 Transfer, De- beam tests, transfer length tests, and end slips observed in velopment, and Splice Length for Strand/Reinforcement in hollow core precast/prestressed concrete members. Source 102 High-Strength Concrete, the Oklahoma Department of Trans- is from a lot of strand delivered to a Midwest precast concrete portation, and NASPA, also known as the Committee of the producer at the same time as another lot from the same American Wire Products Association. Two of the six strand strand manufacturer who experienced excessive slip in hollow sources provided by Russell were actually the same strand core products produced by the precaster. After this precaster source, a fact that was not known before the testing was com- experienced the slip problems, he contracted with Peterman of pleted. This was intended to test the repeatability of the surface KSU to perform concrete pull-out tests. Tests were run on this and chemical test methods. Transfer length and mortar pull- strand with the results shown in Figure B-2. Later, it was re- out test results were provided in tabular form by Russell after ported that this same strand manufacturer was routinely run- the chemical and surface testing had been completed. These

OCR for page 39
44 Figure B-2. Large concrete block pull-out behavior of strands tested at KSU that demonstrated excessive slip in a hollow-core product. In this figure, behavior is compared against that of a well-bonding control. tables are reproduced in Table B-2 and Table B-3. Transfer During the Trial Round, two 4 in. 4 in. 16 ft-prisms length testing had been conducted on only two sources, while were cast for strand designations 102 and 151. The size of the mortar pull-out testing had been conducted on all five prisms was based on two considerations: (1) the cross-sectional sources. Although Russell did not provide a description of his area of the prism was sized to obtain a concrete compressive test methodology, it is believed that the data were obtained as stress of approximately 2000 psi after release of prestress; described in the recently published Master of Science thesis, (2) its length was designed to obtain four transfer lengths per "Assessing the Bond Quality of Prestressing Strands Using specimen--two from the initial release of prestress and two NASP Bond Test" (Chandran 2006). Russell did not supply from saw-cutting the specimen at its mid-point. The prism had the strand tested in Project 12-60 with the lowest bond. to be long enough to assure that the transfer lengths from the four ends would not overlap. Figure B-3 shows a schematic of the transfer length speci- Mechanical Test Methods men, as well as the concrete mixture proportions used for the and Results Trial Round and for Round 1. The cement was a Type III with The original project scope included the development of a a blaine fineness of 564 m2/kg. The coarse aggregate was performance-based test method for use in evaluating strand a siliceous gravel meeting ASTM C33 #67 gradation require- bond. The Screening Round of the experimental program in- ments, and the fine aggregate had a fineness modulus of 2.89. cluded mechanical testing of strand sources using transfer The mixture proportions were determined essentially accord- length, pull-out from concrete (large concrete block pull-out ing to the procedure given in ACI 211.1-91, Standard Practice test [LBPT]), pull out from Portland cement mortar, and pull for Selecting Proportions for Normal, Heavyweight, and Mass out from gypsum plaster-based mortar tests. Each test is pre- Concrete. sented separately in this section. The number of samples of Figure B-4 schematically shows the transfer length specimen each strand source tested by each mechanical test method is mounted in its stressing frame. Figure B-5 shows one such shown in Table B-4 and B-5. frame ready for casting, with prestressing strand tensioned in the form. After the prisms were cast and the formwork was stripped, Whittemore points were attached to both sides of Transfer Length Testing the prisms as shown in Figure B-6. The strain in the beam Transfer length testing was conducted in three rounds, relative to the position along the beam was measured based with its procedure changing slightly for each subsequent on readings taken with a Whittemore gauge (Figure B-7). round of testing. This device measures the change in location of Whittemore

OCR for page 39
Table B-2. NASPA (mortar) pull-out and transfer length test data accompanying strand from OSU.

OCR for page 39
46 Table B-3. Transfer length and beam test data accompanying strand from OSU. Table B-4. Number of samples tested for each mechanical and surface and chemical test method. Strand Pullout Bond Stress Contact Angle pH Source Transfer As- After After pH pH-Fix ID Length Concrete Mortar Hydrocal Rec'd Ca(OH)2 Ignition Meter Duotest 7.5-9.5 Historic Strand KSU-F -- 6 -- -- 1 1 -- -- -- -- KSU-H -- 6 -- -- 2 2 -- -- -- -- SC-F -- 6 -- -- 3 3 -- -- -- -- SC-H -- 6 -- -- 3 3 -- -- -- -- SC-IS -- 6 -- -- 3 3 -- -- -- -- 101 -- 6 -- -- 3 3 -- -- -- -- Recently Manufactured Strand 102 6 6 -- 6 3 3 -- 2 2 2 103 6 6 -- 6 3 3 -- 2 2 2 151 6 6 -- 6 3 3 -- 2 2 2 153 -- 6 -- -- -- 3 -- -- 6 -- OSU Strand Samples 349 -- -- 6 -- -- 3 3 -- 6 -- 548 -- -- 18 -- -- 3 3 -- 6 -- 697 -- -- 11 -- -- 3 3 -- 6 -- 717 -- -- 6 -- -- 3 3 -- 6 -- 478 * -- -- 12 -- -- 3 3 -- 6 -- 960 * -- -- 12 -- -- 3 3 -- 6 -- * Samples designated 478 and 960 were from same source.

OCR for page 39
47 Table B-5. Number of samples tested for each mechanical and surface and chemical test method. Change in Corrosion Strand LAB Potential Corrosion Rate Org. AA and Color. Analysis Source Method Method As- After After Surf. As- After After Ext. ID LOI 1 2 Rec'd Ca(OH)2 Ignition Rough. Rec'd Ca(OH)2 Ignition Res. Water Acid Historic Strand KSU-F 1 2 -- -- -- -- -- 1 -- -- 2 2A 2A A KSU-H 2 2 -- -- -- -- -- 2 -- -- 2 2 2A A SC-F 2 2 -- -- -- -- -- - -- -- 3 3 3A SC-H 2 2 -- -- -- -- -- 2 -- -- 3 3A 3A A SC-IS 2 2 -- -- -- -- -- 2 -- -- 3 2 2A AB 101 2 2 -- 2 -- -- -- 2 -- -- 3 3 3A Recently Manufactured Strand 102 2 2 2 3 1 -- 6 2 1 -- 3 3AB 3A AB 103 2 2 2 3 1 -- 6 2 1 -- 3 3 3A AB 151 2 2 2 3 1 -- 6 2 1 -- 3 3 3A 153 -- -- -- 3 -- 3 12 -- 3 3 3 3 3 OSU Strand Samples 349 3 -- -- 4 -- 3 12 -- 3 3 3 3 3 548 3 -- -- 3 -- 3 12 -- 3 3 3 3 3 697 3 -- -- 3 -- 3 12 -- 3 3 3 3 3 717 3 -- -- 3 -- 3 12 -- 3 3 3 3 3 478 * 3 -- -- 3 -- 3 12 -- 3 3 3 3 3 960 * 3 -- -- 3 -- 3 12 -- 3 3 3 3 3 * Samples designated 478 and 960 were from same source. A Acid wash performed on strand subsequent to water wash. B Both warm- and hot-water washes performed. points to 1/10,000th of an inch. The Whittemore points are three specimens (Figure B-9). For the remaining three cylindrical buttons that contain a central indentation into specimens, release was achieved by gradually heating up the which styli on the gauge are inserted. strand until it failed in tension due to its reduced tensile After the initial readings were recorded, the strand was re- strength at high temperatures. Even using this method, one leased gently, and surface strains were measured. These two of the three remaining prisms split at its end. The strength trial tests resulted in transfer lengths of 20 to 28 in. for each of the concrete at the time of prestress transfer was 4060 psi. source. It was determined that the concrete strength may The strength of the concrete at the time of center was have been too high and the release method too gradual to 4810 psi. provide a true measure of transfer length. The strain versus position was determined after the strand In Round 1, two prisms were cast for each of the three was released. The prisms were then cut in half (Figure B-10), sources of recently manufactured strand (102, 103, and 151); and the strain in the prisms was measured along with the end a total of six stressing frames similar to those needed for the slip (measured at the far ends of the prisms) and strand suck-in Trial Round were used. After casting, in preparation for (measured at the saw-cut ends). measuring the strand end slip, a small notch was made at the To examine the effect of time on transfer length, end slip, and exposed ends of the strand. The distances from this small cut suck-in, these measurements were repeated 28 days, 6 months, to the end of the prism were measured with calipers, in digi- and 22 months after release. Strand suck-in measurements tal photos, and with a steel scale (see Figure B-8). This posi- were abandoned after the 28-day data collection because of tion was later re-measured with all three of these methods their inefficacy. During the initial saw-cutting process, when after release and at later times to determine the end slip of the the saw initially nicked the strand, the individual wires strand. fractured sequentially, leaving an uneven surface from which Initially, the release was achieved by quickly cutting the accurate measurement of the strand displacement proved dif- strand with an acetylene torch similar to what is commonly ficult (see Figure B-11). done in precast concrete plants. However, this resulted in A typical plot of strain versus position on the prisms is splitting and fracturing at the end region of the concrete for given in Figure B-12. This plot shows several sets of strain

OCR for page 39
48 Figure B-3. Transfer length specimen used in the Trial Test and Round 1 Test. data collected over a series of months as detailed above. done by necessity during Round 1, because releasing the Additionally, this plot shows lines drawn at 100% of the strand in this way realistically imitates what is done at a plateau strain for each measurement. The transfer length typical precaster's yard. To prevent end-splitting, specimen measured in this test is defined as the intersection of the strain cross-sections were increased to 4.5 in. 4.5 in., which re- on this plateau and the linear extension of the slope from the duced the compressive stress on the concrete to 1500 psi. transfer region. The Round 2 specimen and stressing frame are presented Because of the problems experienced with this first set of schematically in Figure B-13 and Figure B-14, respectively. transfer length specimens, a second set was fabricated and Figure B-13 also lists the concrete mixture proportions for tested (Round 2). In Round 2, the use of saw cutting to meas- Round 2. The strength of the concrete at the time of prestress ure strand suck-in was omitted due to the previously men- transfer was 4380 psi. tioned ineffectiveness of this method. End slip measurements Because the center saw cut was eliminated, it was possible for this round of testing were continued using the scale and to cast two 8-ft long specimens end-to-end within the fabri- photos. End slips were also measured using a depth gauge. cation frame. A space was left between the specimens using a As a final change, in Round 2, the cross-sectional dimensions stiff standoff, exposing a portion of the strand. The strand in of the prisms were increased to prevent splitting on rapid re- this segment between the two specimens was torch cut after lease of prestress. Rapid release was not abandoned, as was the initial release (see Figure B-15).

OCR for page 39
49 Figure B-4. Transfer length stressing frame and specimen orientation, Trial Test and Round 1 Test. Two additional issues were experienced in this second based on the measured transfer length and strand tensile stress. round: This approach eliminates complications from strands of vary- ing sizes and varying initial stress conditions. The average 1. While making the notch in the wire for the end slip meas- bond stress, Ut, is calculated as urement on two of the prisms, the individual wire fractured and some loss of prestress occurred. f se Aps Ut = (Eq. 1) 2. It was not initially known that one of the strands was C p Lt 0.52-in. diameter (1/2 in. special) instead of 0.5-in. diameter. Thus, this larger strand had a slightly lower bond stress at where fse is the effective prestress after transfer, Aps is the transfer. cross-sectional area of the strand, Cp is the circumferential perimeter of the strand (4/3 db) and Lt is the transfer length. To account for these issues and enable comparisons of Average bond stress is thus dependent both on effective pre- stress transfer behavior between the tested strand sources, the stress as well as transfer length for a given strand geometry. average bond stress over the transfer length was calculated For the specimens from the two rounds of testing, a transfer

OCR for page 39
112 0.700 S=0.0971930 Average Mortar Bond Stress at 0.1-in. Slip 0.600 0.500 0.400 (ksi) 0.300 y = 1.7425x + 0.7659 R2 = 0.6823 0.200 0.100 0.000 -0.4 -0.3 -0.3 -0.2 -0.2 -0.1 -0.1 0.0 Change in Corrosion Potential (V) as-Received Figure B-100. Fitted line plot for mortar pull-out stress versus the change in corrosion potential as-received. Table provides Standard Error (S), R2, and R2 adjusted. The Finally, the bottom table provides the estimates for the multiple R is just the square root of R2 and is sometimes called coefficients (^ = 0.7659 and ^ = 1.7432). The P-values in the 0 1 the correlation. bottom table are the result of statistical tests that test to see if The ANOVA (Analysis of Variance) analysis performed by the true intercept, 0, and the true slope, 1, are equal to zero. Excel includes a statistical test to determine if the amount of The low P-values indicate that neither the slope nor the in- variability explained by the fitted model is significantly more tercept is likely to be zero. For the intercept, this test is not than would be expected from a model fitted to random data very interesting, but if the slope is zero, that would indicate from a normal distribution. All the entries in the ANOVA table no linear relationship between the pull-out stress and the are intermediate values for calculating the "Significance F" change in corrosion potential as-received. Notice that the value. The "Significance F" value shows how likely it would P-value for the slope (in the row labeled "Change in Corr. be that random, normally distributed data would fit a linear Pot. As Received") is exactly the same as the "Significance F" model as well as this model fits these data. Here, the proba- value above. For the case of a single predictor variable, these bility is very low (0.0061) indicating that it is very unlikely that two tests are exactly equivalent. In the next section where these data are just random data and much more likely that the multiple predictors are discussed, these tests will no longer data are actually following the linear model. be equivalent. SUMMARY OUTPUT Regression Statistics Multiple R 0.826022976 R Square 0.682313957 Adjusted R Square 0.636930237 Standard Error 0.097193005 Observations 9 ANOVA df SS MS F Significance F Regression 1 0.142022 0.142022 15.03433 0.00607324 Residual 7 0.066125 0.009446 Total 8 0.208147 Standard Coefficients Error t Stat P-value Lower 95% Upper 95% Intercept 0.765888672 0.105371 7.268469 0.000167 0.516724919 1.015052 Change in Corr. Pot. As Received 1.74252891 0.449405 3.877413 0.006073 0.679854919 2.805203 Figure B-101. Excel output for regression with single predictor.

OCR for page 39
113 The last two entries in each row of the bottom table are 95% specified pull-out bond stress is likely to be achieved, the confidence intervals for the intercept and slope. These are threshold on the QC test must be chosen as the value where ranges that 95% of the time will cover the true values 0 and 1. the prediction interval lower bound is equal to the pull-out The most useful piece of information here is similar to the stress threshold. The model y ^ = ^ + ^ x gives an estimate of 0 1 conclusions made earlier and that is that with 95% confi- the average pull-out stress if the pull-out test was actually dence 1, the slope is greater than zero--again indicating a conducted repeatedly on the same source of strand. For a given linear relationship between pull-out stress and change in cor- measurement of the predictor, half of the actual pull-out test rosion potential as-received. results would be expected to fall above this average and half The main reason for developing this regression is to allow would fall below. The distribution of individual pull-out obser- for prediction of the pull-out stress by measuring the change vations about that average pull-out stress is the basis for the in corrosion potential. However, entering the measured prediction interval, which is calculated based on the variabil- change in corrosion potential into the prediction formula ity in the data used for the regression. y^ = ^ + ^ x gives the average estimated pull-out stress and This concept is demonstrated graphically in Figure B-102, 0 1 does not account for variation that is bound to occur in the which shows the prediction interval lower bound plotted test results or uncertainty in the regression model. This vari- along with the regression line, and data for the mortar pull ation is evidenced by the fact that all the points used to de- out plotted versus the change in corrosion potential. If a spec- velop the regression did not fall on the line, that is the R2 value ified threshold on mortar pull out is defined as 0.313 ksi, the was not 100%. Instead, what is needed to interpret and prac- threshold on the corrosion potential is the value where the tically apply a change in corrosion potential test result is the pull-out threshold and the curve delineating the lower bound computation of a lower bound on the interval that, with 90% of the prediction interval intersect, shown by the red lines in confidence, includes the pull-out stress for a strand sample the plot. In this case, the threshold would be approximately with that change in potential test result. This type of interval -0.175 V. is known as a one-sided prediction interval and is a standard Unfortunately, Excel does not provide prediction intervals as part of regression theory and practice. A two-sided prediction a part of its standard output. However, a formula is provided interval is used when both an upper and a lower bound are in (Eq. 4) that allows for the calculation of a 90% prediction required. The one-sided prediction interval will be focused interval lower bound for pull-out stress of a new strand for the on here. predictor. In the example, this is a prediction interval for pull- The prediction interval concept is a necessary part of the out stress with a measured value of change in corrosion po- development of acceptance/rejection thresholds for the rec- tential as received. Of course, this prediction interval assumes ommended QC tests, since, to conservatively verify that a that the model in (Eq. 3) is the correct model. 0.700 Mortar Pull-out 0.1-in Slip Stress (ksi) 0.600 y = 1.7413x + 0.7656 R2 = 0.6818 0.500 Specified Pull- 0.400 Out Stress Threshold 0.300 Change in Corr. Pot. (V) 0.200 Prediction Interval Lower Bound Threshold on Change in Corr. Pot. (V) 0.100 0.000 -0.35 -0.3 -0.25 -0.2 -0.15 -0.1 -0.05 0 Change in Corr. Pot. (V) Change in Corr Pot. Where Pull-Out Threshold Intersects Prediction Interval Figure B-102. Threshold determination using the prediction interval.

OCR for page 39
114 Prediction Interval (x 0) all the values of the predictor variable that were used in the re- 1 (x 0 - x )2 gression (the prime indicates the transpose of the vector, so xD =^ +^ x -t 0 1 0 .90 ,n- p S 1 + + (Eq. 4) is defined as a column vector), n Sxx Each of the variables in (Eq. 4) is explained in the following x D = (-0.289, -0.080, -0.154, -0.241, -0.272, -0.211, -0.267 table. To help with the calculations below, the vector contains - 0.172, -0.322). Explanation of Variables in Eq. 4 Variable Description Numerical Value in This Example ^ The estimate of the intercept. 0.7658 0 ^ The estimate of the slope. 1.7425 1 S The standard deviation of the observations from the 0.097193 fitted line. x0 The value of x for which y is to be predicted. Any value of corrosion potential for which a prediction is needed. For example, x0 = -0.1. x The average of all the x values used in the regression. x = -0.228 Sxx This is a measure of the spread in the x values in the 0.04683 regression. It is defined as Sxx = x D x D - nx 2 n The total number of observations in the regression. 9 p The total number of 's in the model (0, 1) 2 t.90,n-p The 90th percentile of the t-distribution with n-p 1.42 degrees of freedom. See the table below for approximate for n-p = 7 values. t-distribution for other confidence levels are available in most standard statistics textbooks and using Excel's TINV() function. Approximate Values of the 90th Percentile of the t-Distribution n-p t.90,n-p 1 3.08 2 1.89 3 1.64 4 1.53 5 1.48 67 1.42 89 1.39 1013 1.36 1427 1.33 >27 1.30

OCR for page 39
115 Following (Eq. 4), the 90% prediction interval for a measures and pull-out stress is not conclusive at the chosen new strand sample that has a change in corrosion potential confidence level. as-received of x0 = -0.1 V, would be If this multiple regression model were to be used for pre- diction of the pull-out stress, a prediction interval is still = 0.7658 + 1.7425(-0.1) needed to determine thresholds. Determining the predic- tion interval for models based on multiple predictors is 1 (-0.1 - (-0.228))2 - 1.42(0.097193) 1 + + possible; however, it is somewhat more complicated and 9 0.04683 = 0.425 involves the matrix manipulations demonstrated by the fol- lowing example. This means that for a measured change in corrosion For this example, the data used to develop the model by potential as-received of -0.100 V, the mortar pull-out stress regression are as follows: at 0.1-in. slip would be expected, with 90% confidence, to be greater than 0.425 ksi. Change in Mortar Pull- For the purpose of determining thresholds on other surface Weight Contact Corrosion Out Stress and chemical tests, the prediction interval over the range of Loss on Angle after Potential at 0.1-in. the measured responses is needed. This is calculated by vary- Ignition Lime Dip As-Received Slip (ksi) ing x0 in Eq. 4. x1 x2 x3 y 0.139 87 -0.289 0.156 Regression with Multiple Predictors 0.059 79 -0.08 0.623 Often the response of interest depends on more than one 0.036 68 -0.154 0.606 predictor variable. In that case, additional terms are added to 0.086 94 -0.241 0.206 the regression model as shown in Eq. 5 for k predictors. 0.041 73 -0.272 0.409 0.045 76 -0.211 0.409 0.051 87 -0.267 0.315 y = 0 + 1 x1 + 2 x 2 + + k x k + (Eq. 5) -0.021 79 -0.172 0.397 0.002 98 -0.322 0.273 When regression is performed with multiple predictors, the regression output is very similar to the regression output for a single predictor. Again, the R2 and R2 adjusted can be Let the matrix X be defined as a column of ones (repre- calculated and should be interpreted as measures of how senting the coefficient to be multiplied by the intercept, 0) well the model fits the data. The difference between the two and then a column of levels for each of the other predictors in statistics is that, since adding predictors makes the model more each observation. flexible and thus better able to fit the data, the R2 adjusted measure includes a penalty for additional predictors in the 1 0.139 87 -0.289 model. Thus, the goal of multiple regression is to find a model 1 0.059 79 -0.080 that has a relatively high R2 value with as few predictors as 1 0.036 68 -0.154 possible, and maximizing R2 adjusted accomplishes this goal. 1 0.086 94 -0.241 However, if too many predictors are put into the model, there X = 1 0.041 73 -0.272 will be very few degrees of freedom (n-p) for estimating the 1 0.045 76 -0.211 error variance (S). Again the goal is to have low values of the 1 0.051 87 -0.267 1 -0.021 -0.172 F-significance and low P-values for each of the slope estimates 79 (i.e., coefficients). An Excel regression output for a multiple 1 0.002 98 -0.322 regression is shown in Figure B-103, and it can be seen that this regression is a better model for the data than the single The 90% prediction interval is then given by: predictor regression shown above, as measured by R2 and R2 adjusted. However, two of the slope estimates (Weight LOI and Contact Angle After Dip) have P-values that are rel- Prediction Interval (x1 , x 2 , . . . , x k ) = ^ + ^ x + ^ x ^ x -t .90 ,n- p S x (X X ) x , -1 (Eq. 6) atively large. Notice that the 95% confidence intervals of the 0 1 1 2 2 k k slopes for these two factors contain zero; this indicates that the data are not conclusively supporting that these slopes are where each of the variables in (Eq. 6) is explained in the fol- different than zero. Thus, the relationship between these lowing table.

OCR for page 39
116 Explanation of Variables in Eq. 6 Variable Description Numerical Value in This Example ^ The estimate of the intercept. 0.7658 0 ^ The estimate of the slope for predictor variable i. 1.7425 i X A vector of levels for each predictor at which y is to 1 For example, x 0 = be predicted (including a 1 for the intercept). 0.04 78 -0.250 n The total number of observations in the regression. 9 k The number of predictors in the model. 3 p=k+1 The total number of 's in the model. 4 t.90,n-p The 90th percentile of the t-distribution with n-p 1.48 for n-p = 5 degrees of freedom. See the table above for approximate values. t-distribution for other confidence levels are available in most standard statistics textbooks and using Excel's TINV() function. 1 Unfortunately, the prediction intervals based on regression with multiple predictors cannot be plotted in two dimensions The 90% prediction interval lower bound for x 0 = 0.04 78 as was done in Figure B-103, for the single predictor example. -0.250 The predicted pull-out stress (y in Eq. 5) is not uniquely de- is 0.252 ksi. termined by a single combination of predictors (x1, x2, . . .), This means that for a strand source with measured LOI of but can be found based on numerous combinations. However, 0.04 mg/cm2, a contact angle after lime dip of 78, and a the prediction interval for the pull-out stress will be different change in corrosion potential as-received of -0.250 V, the depending on the specific combination of predictors that mortar pull-out stress at 0.1-in. slip would be expected, is used. That means that when multiple regression is used with 90% confidence, to be greater than 0.252 ksi. to improve the predictive ability of the model, a single SUMMARY OU TPUT Regression Statistics Multiple R 0.922308764 R Square 0.850653457 Adjusted R Square 0.761045531 Standard Error 0.078849246 Observations 9 ANOVA df SS MS F Significance F Regression 3 0.177061 0.05902 9.49306 0.016594131 Residual 5 0.031086 0.006217 Total 8 0.208147 Standard Coefficients Error t Stat P-value Lower 95% Upper 95% Intercept 1.202523363 0.243392 4.940688 0.00432 0.576864667 1.828182 Weight Loss on Ignition -0.84478002 0.617114 -1.368919 0.229318 -2.431123313 0.741563 Contact Angle After Lime Dip -0.00632965 0.003492 -1.812644 0.129631 -0.015305971 0.002647 Change in Corr. Pot. As Received 1.179495734 0.450947 2.6156 0.047349 0.020300812 2.338691 Figure B-103. Excel output for regression with multiple predictors.

OCR for page 39
117 Table B-38. Regression coefficients for single-predictor models. Coefficient of Predictor Constant Coefficient Determination (x0) ( ) (R2) Weight Loss on Ignition (mg/cm2) 0.445 -1.403 0.16 Contact Angle after Lime Dip () 1.393 -0.012 0.57 Change in Corrosion Potential After 6 h(V)--As-Received 0.766 1.741 0.68 Extracted Organic Residue (mg/cm2) 0.453 -1.752 0.12 Extracted Organic Residue (mg/cm2)--Stearate only 0.436 -1.943 0.63 threshold cannot be defined. Instead, for a specific set of pre- Thresholds Based on Regression dictors, a new prediction interval must be calculated based on with Single Predictor the set of data used to develop a regression model. The lower The test methods that were recommended for inclusion are: bound of the newly calculated prediction interval must then be compared with the specified pull-out threshold. For the Weight LOI, example calculation performed above, the lower bound is Contact Angle Measurement after Lime Dip, 0.252 ksi. This is lower than the specified pull-out threshold Change in Corrosion Potential, of 0.313 ksi, so it cannot be predicted that this source of Organic Residue Extraction. strand will exceed that pull-out stress 90% of the time (the de- fined confidence level). The framework for completing this cal- The efforts made to define thresholds for each of these culation and comparison is given in the Microsoft Excel methods were based on single variable linear regressions and spreadsheet developed in this study. are described individually below. The results are summarized in Table B-38. Selection of Confidence Level Weight Loss on Ignition (LOI)--The prediction interval For the threshold determinations performed based on for LOI with a one-sided confidence level of 90% is shown in the data collected in this study, the confidence level was taken Figure B-104. As can be seen in this figure, the prediction in- as 90%. This means that for a given surface and chemical test terval does not exceed 0.313 ksi anywhere over the range of result, 10% of the pull-out results would be expected to fall test results observed in this study. For that reason, no threshold below that prediction interval. This confidence level is lower can be determined. than the 95% confidence interval that is most commonly used as the basis for probabilistic design in structural engineering Contact Angle Measurement after Lime Dip--The pre- analysis. However, using a confidence level as high as 95% diction interval for Contact Angle with a one-sided confidence makes determination of the thresholds for the surface and level of 90% is shown in Figure B-105. As can be seen in this chemical tests very conservative. figure, this prediction interval exceeds 0.313 ksi when the 0.700 Mortar Pull-Out 0.1-in. Slip Stress (ksi) 0.600 Loss on Ignition (mg/cm2) 0.500 Prediction Interval Lower Bound Threshold on Loss on Ignition (mg/cm2) 0.400 0.300 0.200 0.100 y = -1.4031x + 0.4454 R2 = 0.1595 0.000 -0.100 -0.04 -0.02 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 Loss on Ignition (mg/cm2) Figure B-104. Prediction interval (confidence level 90%) for Weight Loss on Ignition. Threshold not possible.

OCR for page 39
118 0.700 Mortar Pull-Out 0.1-in. Slip Stress (ksi) 0.600 0.500 0.400 Contact Angle After Lime () Prediction Interval Lower Bound 0.300 Threshold on Contact Angle After Lime () 0.200 y = -0.0123x + 1.3929 0.100 2 R = 0.5701 0.000 -0.100 0 20 40 60 80 100 120 Contact Angle after Lime () Figure B-105. Prediction interval (confidence level 90%) for Contact Angle after Lime Dip. contact angle is less than 73. Therefore, based on these data a Change in Corrosion Potential of -0.175 V or more (less and the NASPA defined threshold on mortar pull-out stress negative) is recommended to give a good confidence of ade- at 0.1-in. slip, a Contact Measurement after Lime Dip of 73 quate bond. or lower is recommended to give a good (90%) confidence of adequate bond. This test must be run on recently manufac- Organic Residue Extraction--The prediction interval tured strand with no surface weathering or rust (i.e., bright for organic residue extraction with a one-sided confidence strand). level of 90% is shown in Figure B-107. As can be seen in this figure, the prediction interval does not exceed 0.313 ksi Change in Corrosion Potential--The prediction interval anywhere over the range of test results observed in this for Change in Corrosion Potential with a one-sided confi- study. For that reason, no threshold can be determined. dence level of 90% is shown in Figure B-106. As can be seen A similar analysis was attempted considering only those in this figure, this prediction interval exceeds 0.313 ksi when sources with organic residue that the FTIR analyses in- the change in the corrosion potential is less negative than dicated was primarily stearate. This was done to eliminate -0.175 V. Therefore, based on these data and the NASPA potentially confounding influences of non-stearate-based defined threshold on mortar pull out 0.1-in. slip stress, lubricants. The prediction interval for this stearate residue 0.700 Mortar Pull-Out 0.1-in. Slip Stress (ksi) 0.600 y = 1.7413x + 0.7656 2 R = 0.6818 0.500 0.400 0.300 Change in Corr. Pot. (V) 0.200 Prediction Interval Lower Bound Threshold on Change in Corr. Pot. (V) 0.100 0.000 -0.35 -0.3 -0.25 -0.2 -0.15 -0.1 -0.05 0 Change in Corr. Pot. (V) Figure B-106. Prediction interval (confidence level 90%) for Change in Corrosion Potential.

OCR for page 39
119 0.700 Mortar Pull-Out 0.1-in. Slip Stress (ksi) Organic Residue (mg/cm2) 0.600 Prediction Interval Lower Bound 0.500 Threshold on Organic Residue (mg/cm2) 0.400 0.300 0.200 y = -1.7515x + 0.453 2 R = 0.1241 0.100 0.000 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 Organic Residue (mg/cm2) Figure B-107. Prediction interval (confidence level 90%) for Organic Residue. Threshold not possible. with a one-sided confidence level of 90% is shown in Fig- showed the best correlation, based on the adjusted coefficient ure B-108. As can be seen in this figure, the Coefficient of of determination (R2 adj.), were: Determination is higher, but the prediction interval still does not exceed 0.313 ksi anywhere over the range of test Contact Angle Measurement after Lime Dip & Change in results observed in this study, and no threshold can be Corrosion Potential, determined. Contact Angle Measurement after Lime Dip & Organic Residue Extraction (100% stearate only), Weight Loss on Ignition (LOI) & Contact Angle Measure- Thresholds Based on Regression ment after Lime Dip & Change in Corrosion Potential. with Multiple Predictors An attempt was also made to look for linear combinations Note that for multiple-predictor regression, a larger num- of multiple results obtained from all the evaluated methods ber of variables will increase the R2. Therefore, the adjusted R2 that would correlate with bond performance. While numerous statistic, which accounts for the number of degrees of freedom combinations were examined, the three combinations that in the data set, was calculated as a means to compensate for 0.450 Organic Residue (mg/cm2) Mortar Pull-Out 0.1-in. Slip Stress (ksi) 0.400 Prediction Interval Lower Bound 0.350 Threshold on Extracted Organic Residue (mg/cm2) 0.300 0.250 0.200 y= -1.9426x + 0.4355 0.150 R2 = 0.6282 0.100 0.050 0.000 0.000 0.020 0.040 0.060 0.080 0.100 0.120 0.140 Extracted Organic Residue (mg/cm2) Figure B-108. Prediction interval (confidence level 90%) for Organic Residue when FTIR analysis indicates organic residue is primarily stearate. Threshold not possible.

OCR for page 39
120 Table B-39. Regression coefficients for model based Table B-40. Regression coefficients for model based on Contact Angle Measurement after Lime Dip & on Contact Angle Measurement after Lime Dip & Change in Corrosion Potential. Organic Residue Extraction (100% stearate only). Predictor Coefficient Predictor Coefficient Constant 1.209 Constant 0.864 Contact Angle after Lime Dip () -0.007 Contact Angle after Lime Dip () -0.006 Change in Corrosion Potential After 6 h (V)--As-Received 1.233 Extracted Organic Residue (mg/cm2) -1.093 Adjusted Coefficient of Determination (R2 adj.) 0.727 Adjusted Coefficient of Determination (R2 adj.) 0.976 this potentially misleading effect. The regression coefficients Table B-41. Regression coefficients for model for these three models and the R2 adj. are given in Table B-39 based on Weight Loss on Ignition (LOI), Contact to Table B-41. The R2 adj. values for these combinations were Angle Measurement after Lime Dip & Change high and equal to 0.73, 0.98 and 0.76, respectively. The R2 adj. in Corrosion Potential. for each of these combinations was higher than the R2 for the Predictor Coefficient single-predictor regression models. Constant 1.203 The regression that indicated that the last combination of Weight Loss on Ignition (mg/cm2) -0.846 predictors listed above (Contact Angle Measurement after Contact Angle after Lime Dip () -0.006 Lime Dip & Organic Residue Extraction) was a good predic- Change in Corrosion Potential After 6 h (V)--As-Received 1.178 Adjusted Coefficient of Determination (R2 adj.) 0.761 tor of bond was performed based only on those strand sources that the FTIR analysis of the organic residue identi- fied as being stearate only. This limited the number of data sheet has been developed for this purpose. To give a sense of points used to develop the regression model to five, but was how these multiple regression models might be used, tables done as a means of eliminating potentially confounding in- have been prepared showing the predicted pull out, the lower fluences of non-stearate-based lubricants on the results bound on the prediction interval and the result of a compar- obtained by the contact angle and organic residue extraction ison with the specified mortar pull-out threshold of 0.313 ksi, measurement methods. Given the high level of correlation for each of the three multiple regression models. These are with the multiple regression approach, this model may be shown in Table B-42 to Table B-44. particularly useful in a production setting where the lubricant For example, Table B-42 was developed for the model in use is known. based on Contact Angle Measurement after Lime Dip & The prediction interval cannot be shown in a two- Change in Corrosion Potential. The first row of this table dimensional plot as was done with the single variable models. shows the results of these two individual tests obtained for This is because there are multiple combinations of variables Source 349. Based on the regression model, the predicted that can combine to give the same output. For this reason, mortar pull-out stress at 0.1-in. slip is 0.264 ksi for these a separate prediction interval must be calculated for each two results. The lower bound on the prediction interval for combination of variables. A Microsoft Excel-based spread- that combination of the two test results must be calculated Table B-42. Evaluation of prediction interval for model based on Contact Angle Measurement after Lime Dip & Change in Corrosion Potential. Mortar Pull Out 0.1-in Slip Stress (ksi) Contact Pass/Fail* Pass/Fail* Strand Angle Change in Value Based on Experimentally Lower Bound Based on Source after Corrosion Predicted by Prediction Determined in of Prediction Pull-Out Test ID Lime Potential (V) Regression for Interval from Pull-Out Test Interval Result Dip () QC Results QC Tests 349 87 -0.289 0.156 0.264 0.131 Fails Fails 548 79 -0.080 0.623 0.576 0.420 Passes Passes 697 68 -0.154 0.606 0.559 0.417 Passes Passes 717 94 -0.241 0.206 0.276 0.136 Fails Fails 478 73 -0.272 0.409 0.38 0.232 Fails Passes 960 76 -0.211 0.409 0.435 0.303 Fails Passes 102 87 -0.266 0.315 0.291 0.161 Fails Passes 103 79 -0.172 0.397 0.463 0.331 Passes Passes 151 98 -0.322 0.273 0.149 0.003 Fails Fails * Threshold for passing is 0.313 ksi.

OCR for page 39
121 Table B-43. Evaluation of prediction interval for model based on Contact Angle Measurement after Lime Dip & Organic Residue Extraction (100% stearate only). Mortar Pull Out 0.1-in Slip Stress (ksi) Contact Pass/Fail* Extracted Pass/Fail* Strand Angle Value Based on Organic Experimentally Lower Bound Based on Pull Source after Predicted by Prediction Residue Determined in of Prediction Out Test ID Lime Regression for Interval from (mg/cm2) Pull-Out Test Interval Result Dip () QC Results QC Tests 717 94 0.117 0.206 0.211 0.176 Fails Fails 478 73 0.033 0.409 0.42 0.388 Passes Passes 960 76 0.035 0.409 0.401 0.371 Passes Passes 102 87 0.069 0.315 0.303 0.274 Fails Passes 151 98 0.037 0.273 0.276 0.24 Fails Fails * Threshold for passing is 0.313 ksi. Table B-44. Evaluation of prediction interval for model based on weight loss on ignition, Contact Angle Measurement after Lime Dip & Change in Corrosion Potential. Mortar Pull Out 0.1-in Slip Stress (ksi) Contact Change Pass/Fail* Weight Pass/Fail* Strand Angle in Value Lower Based on Loss on Experimentally Based on Pull Source after Corrosion Predicted by Bound of Prediction Ignition Determined in Out Test ID Lime Potential Regression for Prediction Interval from (mg/cm2) Pull-Out Test Result Dip () (V) QC Results Interval QC Tests 349 0.139 87 -0.289 0.156 0.193 0.044 Fails Fails 548 0.059 79 -0.08 0.623 0.558 0.407 Passes Passes 697 0.036 68 -0.154 0.606 0.56 0.424 Passes Passes 717 0.086 94 -0.241 0.206 0.25 0.114 Fails Fails 478 0.041 73 -0.272 0.409 0.385 0.243 Fails Passes 960 0.045 76 -0.211 0.409 0.435 0.309 Fails Passes 102 0.051 87 -0.267 0.315 0.294 0.169 Fails Passes 103 -0.021 79 -0.172 0.397 0.517 0.378 Passes Passes 151 0.002 98 -0.322 0.273 0.2 0.049 Fails Fails * Threshold for passing is 0.313 ksi. specifically using those values and is 0.131 ksi. Since 0.131 ksi range of strand sources to help establish correlations between is less than the mortar pull-out threshold of 0.313 ksi, this the proposed QC tests methods and bond quality. source fails to meet the minimum required combined Con- To evaluate these test methods, several rounds of evaluation tact Angle Measurement after Lime Dip & Change in Corro- were conducted: screening, correlation, and precision testing. sion Potential performance. For Source 548, the prediction The objective for the Screening Round was to eliminate interval calculated for the specific combination of test results those tests that do not show promise for predicting bond measured for that source is 0.420 ksi, and this source "passes" performance. The Correlation Round included those methods since this value exceeds the threshold. that showed promise in the screening experiments and was conducted to confirm that the selected QC tests correlated with bond performance over a larger sample set and were able Summary to accurately identify good and bad strand. The Precision An experimental program was conducted to evaluate a num- Testing was intended to form the base for precision statements ber of test methods proposed for use as part of a QC program to be included in the published test methods. to evaluate bond of strand. This included limited mechanical The Screening Round was conducted using strand collected testing (pull-out testing from concrete, Portland cement mor- by the project team over the 5 years before this project began. tar, and gypsum plaster-based mortar) and extensive surface Obtaining additional strand with a range of bonding qualities and chemical testing (Contact Angle, Examination Under for the purposes of the Correlation Round directly from strand UV light, pH, LOI, Loss in Alkali Bath, Change in Corrosion suppliers proved difficult. Sections of strand were provided by Potential, Corrosion Rate, Surface Roughness, Organic Residue Bruce Russell of OSU, who had previously conducted mortar Extraction/FTIR Analysis, and Elemental Analysis). These pull-out tests on these sources of strand in work performed tests, as well as transfer length tests, have been conducted on a for the NCHRP Project No. 12-60 Transfer, Development,

OCR for page 39
122 and Splice Length for Strand/Reinforcement in High-Strength for each combination of test results. A computational tool in Concrete; the Oklahoma Department of Transportation; and the form of a Microsoft Excel spreadsheet has been developed NASPA, also known as the Committee of the American Wire for this purpose. Products Association [AWPA]). Although pull-out testing from concrete appears to corre- References late best with transfer length, the most reliable and realistic measure of bond performance, the Correlation Round of this Chandran, K. (2006). Assessing the Bond Quality of Prestressing Strands test program had to be based on available mortar pull-out Using NASP Bond Test, Master of Science Thesis, Oklahoma State University, Stillwater, OK. results provided by Russell from the NCHRP 12-60 program. Hyett, A.J., Dube, S., and Bawden, W.F. (1994, November). "Laboratory The four test methods that showed the best correlation Bond Strength Testing of 0.6" 7-Wire Strand from 7 Different with bond and that are recommended for inclusion in future Manufacturers." Final Report. Department of Mining Engineering, QC programs are: Queen's University, Kingston, Ontario. Lane, S.N. (1998, December). "A New Development Length Equation for Pretensioned Strands in Bridge Beams and Piles," Report No. FHWA- 1. Weight Loss on Ignition (LOI) of Strand (QC-I), RD-98-116, Federal Highway Administration, Structures Div., 2. Contact Angle Measurement of a Water Droplet on a McLean, VA. Strand Surface (QC-I), Logan, D.R. (1997 March-April). "Acceptance Criteria for Bond Qual- 3. Change in Corrosion Potential of Strand (QC-I), and ity of Strand for Pretensioned Prestressed Concrete Applications," 4. Organic Residue Extraction with FTIR Analysis (QC-II). PCI Journal, pp. 5290. Mitchell. D., Cook. W. D., Khan. A. A., and Tham, T. (1993, May-June). "Influence of High Strength Concrete on Transfer and Develop- The QC tests have been divided into two categories, de- ment Length of Pretensioned Strand," PCI Journal, Vol. 38. No. 3., pending on the complexity and time required to conduct the pp. 5266. tests: Level I (QC-I) and Level II (QC-II) tests. The QC level Moustafa, S. (1974). "Pull-Out Strength of Strand Lifting Loops." Tech- is shown in the list above. nical Bulletin 74-B5, Tacoma, WA: Concrete Technology Associates. Thresholds for these QC tests have been developed where Perenchio, W.F., Fraczek, J., and Pfeiffer, D.W. (1989). Corrosion Protection of Prestressing Systems in Concrete Bridges. NCHRP possible based on prediction intervals for the regression Report No. 313, Transportation Research Board, National Research calculated from the available data and a minimum criterion Council, Washington, D.C. on the mortar pull-out stress adopted by NASPA. Peterman, R.J. (2007, May-June). "The Effects of As-Cast Depth and Regression with multiple predictors has also been performed Concrete Fluidity on Strand Bond," PCI Journal, pp. 72101. to see if results of these methods can be combined to better Post-Tensioning Institute. (1996). Recommendations for Prestressed Rock predict bond. The three combinations that showed the best and Soil Anchors, 3rd Ed., Phoenix, AZ. Rose, D.R., and Russell, B.W. (1997, July/August). "Investigation of Stan- correlation, based on the adjusted coefficient of determination dardized Tests to Measure the Bond Performance of Prestressing (R2 adj.), were: Strand," PCI Journal, Vol. 42, No. 4, pp. 5680. Russell, B.W., and Paulsgrove, G.A. (1999). "NASP Strand Bond Testing 1. Weight Loss on Ignition (LOI) & Contact Angle Measure- Round Two--Assessing Repeatability and Reproducibility of the ment after Lime & Change in Corrosion Potential, Moustafa Test, the PTI Bond Test and the NASP Bond Test." Final 2. Contact Angle Measurement after Lime & Change in Cor- Report 99-04. The University of Oklahoma, Fears Structural Engi- neering Laboratory, Norman, OK. rosion Potential, and Russell, B.W. (2001). "Final Report--NASP Round III Strand Bond 3. Contact Angle Measurement After Lime & Organic Residue Testing," Okalahoma State University, Stillwater, OK. Extraction (100% stearate only). Russell, B.W. (2006, June). "Final Report--NASP Round IV Strand Bond Testing" Okalahoma State University, Stillwater, OK. The adjusted coefficients of determination for each of these Stocker, M.F., and Sozen, M.A. (1971). "Investigation of Prestressed combinations were higher than the coefficients of determina- Reinforced Concrete for Highway Bridges, Part V: Bond Character- istics of Prestressing Strand," Engineering Experiment Station 503, tion for the single-predictor regression models. Thresholds for The University of Illinois, Urbana-Champaign, IL. multiple-predictor regressions cannot be determined using the Tabatabai, H., and Dickson, T.J. (1993, Nov-Dec). "The History of the same procedure used for single-predictor regressions. Instead, Pretensioned Strand Development Length Equation," PCI Journal, the lower bound on the prediction interval must be calculated Vol. 38, No. 6, pp. 6475.