Analysis of the Sensitivity of a Concrete Virtual Model to Different Material Input Parameters

MISSING IMAGE

Material Information

Title:
Analysis of the Sensitivity of a Concrete Virtual Model to Different Material Input Parameters
Physical Description:
1 online resource (134 p.)
Language:
english
Creator:
Watts, Benjamin E
Publisher:
University of Florida
Place of Publication:
Gainesville, Fla.
Publication Date:

Thesis/Dissertation Information

Degree:
Master's ( M.S.)
Degree Grantor:
University of Florida
Degree Disciplines:
Civil Engineering, Civil and Coastal Engineering
Committee Chair:
FERRARO,CHRISTOPHER CHARLES
Committee Co-Chair:
HAMILTON,HOMER ROBERT,III
Committee Members:
GURLEY,KURTIS R

Subjects

Subjects / Keywords:
cement -- concrete -- modeling
Civil and Coastal Engineering -- Dissertations, Academic -- UF
Genre:
Civil Engineering thesis, M.S.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract:
The development, testing, and certification of new concrete mix designs is an expensive and time consuming aspect of the concrete industry. A software package, labeled the Virtual Concrete and Cement Testing Laboratory (VCCTL) has been developed by the National Institute of Standard sand Technology as a tool to predict the performance of concrete mixes quickly using computer simulation of the hydration behavior of concrete. This software requires specific data on the materials being simulated, such as cement phase volume and surface area fraction, particle size distribution, gradation,density and other properties in order to accurately perform these predictions. A two phase testing program was implemented in conjunction with the Florida Department of Transportation to evaluate the effectiveness of the VCCTL for the prediction of concrete performance. The techniques required to characterize Portland cement were developed and implemented to provide input data for the VCCTL. The resulting virtual materials were simulated, and a testing program was performed on physical specimens to evaluate the accuracy of those simulations. Experimental values of compressive strength, elastic modulus, and time of set for different mixtures were compared to the values predicted by the simulation of these mixtures within the software.
General Note:
In the series University of Florida Digital Collections.
General Note:
Includes vita.
Bibliography:
Includes bibliographical references.
Source of Description:
Description based on online resource; title from PDF title page.
Source of Description:
This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility:
by Benjamin E Watts.
Thesis:
Thesis (M.S.)--University of Florida, 2013.
Local:
Adviser: FERRARO,CHRISTOPHER CHARLES.
Local:
Co-adviser: HAMILTON,HOMER ROBERT,III.

Record Information

Source Institution:
UFRGP
Rights Management:
Applicable rights reserved.
Classification:
lcc - LD1780 2013
System ID:
UFE0046407:00001


This item is only available as the following downloads:


Full Text

PAGE 1

1 ANALYSIS OF THE SENSITIVITY OF A CONCRETE VIRTUAL MODEL TO DIFFERENT MATERIAL INPUT PARAMETERS By BENJAMIN E. WATTS A THESIS SUBMITTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIR EMENTS FOR THE DEGREE OF MASTER OF SCIENCE UNIVERSITY OF FLORIDA 2013

PAGE 2

2 2013 Benjamin E. Watts

PAGE 3

3 ACKNOWLEDGMENTS I would like to thank Dr. Chris Ferraro for the opportunity to pursue an advanced degree in cementitious research, as well as for all his help editing and assembling this document. I would also like to thank Dr. Hamilton and Dr. Gurley for being members of my committee and participating in my defense. I would also like to acknowledge the contributions of Dale Deford, Mike Bergin, Richard DeLorenzo, Pat Carlton, Thomas Frank, Jose Armenteros, and Patrick Gallagher at the FDOT. Their assistance and advice was invaluable. Much of this research would have been impossible without the help and cooperation of Dr. Amelia Dempere and the other staff at the MAIC. The advice and assistance of Jeff Bullard, Ed Garboczi, and Paul Stutzman, and Dale Bentz from NIST with t he VCCTL was also invaluable. I appreciate the assistance of a number of my fellow students with this research as well, including Justin Henika, Jordan Nelson, Jerry Paris, Michael Perry, and Ashley Kerr. Finally, I would like to thank my family, for their universal support and unwavering desire to enable their children to have better lives and more opportunities than they had themselves. For them I am very grateful.

PAGE 4

4 TABLE OF CONTENTS P age ACKNOWLEDGMENTS ................................ ................................ ................................ ............... 3 LIST OF TABLES ................................ ................................ ................................ ........................... 7 LIST OF FIGURES ................................ ................................ ................................ ......................... 8 ABSTRACT ................................ ................................ ................................ ................................ ... 12 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .................. 13 Background ................................ ................................ ................................ ............................. 13 Model Function ................................ ................................ ................................ ....................... 13 Research Requirements ................................ ................................ ................................ .......... 14 Research Objectives ................................ ................................ ................................ ................ 14 Hypothesis ................................ ................................ ................................ .............................. 15 Significance of Research ................................ ................................ ................................ ........ 15 Outline of Thesis ................................ ................................ ................................ ..................... 15 2 LITERATURE REVIEW ................................ ................................ ................................ ....... 17 Portland Cement Hydration ................................ ................................ ................................ .... 17 Ce ment Heat of Hydration ................................ ................................ ................................ ...... 18 Admixtures ................................ ................................ ................................ ............................. 19 Strength of Concrete ................................ ................................ ................................ ............... 21 Scanning Electron Microscopy ................................ ................................ ............................... 22 Computer Modeling ................................ ................................ ................................ ................ 25 Cementitio us Simulation ................................ ................................ ................................ ........ 27 Background ................................ ................................ ................................ ...................... 27 History ................................ ................................ ................................ ............................. 27 CEMHYD3D ................................ ................................ ................................ ................... 33 Existing Research using the VCCTL ................................ ................................ ...................... 35 Current Limitations of the VCCTL ................................ ................................ ........................ 37 Other Cementitious Modeling Packages ................................ ................................ ................ 38 3 RESEARCH APPROACH ................................ ................................ ................................ ..... 40 4 VCCTL MATERIAL INPUTS ................................ ................................ ............................... 42 Over view ................................ ................................ ................................ ................................ 42 Inputs for Portland Cement ................................ ................................ ................................ ..... 43 Microstructural Inputs ................................ ................................ ................................ ..... 44 Gypsum Phase Mass Fractions ................................ ................................ ........................ 44

PAGE 5

5 Cement Phase Volume and Surface Area Fractions, Two Point Correlation Functions ................................ ................................ ................................ ...................... 44 Cement Particle Size Distribution ................................ ................................ ................... 46 Cement Particle Shape Data ................................ ................................ ............................ 47 Cement Particle Dispersion ................................ ................................ ............................. 48 Binder System Size ................................ ................................ ................................ .......... 48 Hydration Modeling Inputs ................................ ................................ ................................ ..... 48 Isothermal Conduction Calorimetry ASTM C 1702 ................................ .................... 49 Curing Conditions ................................ ................................ ................................ ........... 53 Aggregate Input Data ................................ ................................ ................................ .............. 54 5 SEM MICROANALYSIS ................................ ................................ ................................ ...... 56 Overview ................................ ................................ ................................ ................................ 56 Sample Preparation ................................ ................................ ................................ ................. 56 Improvements ................................ ................................ ................................ .................. 57 Backscatter Elec tron and X ray Map Image Acquisition ................................ ....................... 62 Image Processing ................................ ................................ ................................ .................... 66 Creation of Segmented Image ................................ ................................ ......................... 69 Automated Cement Characterization ................................ ................................ ...................... 73 6 VCCTL OUTPUT DATA ................................ ................................ ................................ ...... 79 Overview ................................ ................................ ................................ ................................ 79 Continuous Measurements ................................ ................................ ................................ ...... 80 Periodic Measurements ................................ ................................ ................................ ........... 83 Elastic Modulus ................................ ................................ ................................ ............... 84 Hydrated Cem ent Microstructure Modulus ................................ ................................ ..... 84 Concrete and Mortar Modulus ................................ ................................ ......................... 85 Compressive Strength ................................ ................................ ................................ ...... 86 7 PHYSIC AL TEST PROGRAM ................................ ................................ ............................. 90 Physical Testing Program ................................ ................................ ................................ ....... 90 Isothermal Calorimetry ................................ ................................ ................................ ........... 94 8 COMPARISON OF PHYSICAL TEST DATA TO MODEL OUTPUTS ............................ 95 Experimental Design ................................ ................................ ................................ .............. 95 Simulation Procedures ................................ ................................ ................................ ............ 97 Results ................................ ................................ ................................ ................................ ..... 98 Water to cement Ratio Study ................................ ................................ .......................... 98 Input Sensitivity Study ................................ ................................ ................................ .. 105 Admixture Set Time Study ................................ ................................ ............................ 109 Discussion of Results ................................ ................................ ................................ ............ 117 Modulus and Strength ................................ ................................ ................................ .... 117 Model Sensitivity ................................ ................................ ................................ ........... 122 Admixture Influence on Set Time ................................ ................................ ................. 12 3

PAGE 6

6 9 CONCLUSIONS ................................ ................................ ................................ .................. 124 10 RECOMMENDATIONS FOR FUTURE RESEARCH ................................ ...................... 126 APPENDIX A MATHEMATICAL EQUATIONS REFERENCE FOR THE CALCULATION OF ELASTIC MODULI USING DIFFERENTIAL EFFECTIVE MEDIUM THEORY ......... 128 B SEM FIELD RAW DATA ................................ ................................ ................................ ... 130 LIST OF REFERENCES ................................ ................................ ................................ ............. 131 BIOGRAPH ICAL SKETCH ................................ ................................ ................................ ....... 134

PAGE 7

7 LIST OF TABLES Table P age 5 1 Grinding and Polishing Procedure ................................ ................................ ..................... 61 5 2 Volume and Surface Area Fraction Averages ................................ ................................ ... 72 5 3 Standard Deviations of Volume and Surface Area Fraction Measurements ..................... 72 5 4 Average Difference Between two Image Sets ................................ ................................ ... 74 5 5 XRD Cement Phase Fraction vs. SEM Microanalysis Measurements .............................. 77 7 1 Mixture Design Summary ................................ ................................ ................................ .. 90 7 2 Summary of Test Results for Fresh Concrete ................................ ................................ .... 91 7 3 Tests and Ages ................................ ................................ ................................ ................... 94 B 1 Fiel ds Acquired At RJ Lee ................................ ................................ ............................... 130 B 2 Fields Acquired At UF ................................ ................................ ................................ ..... 130

PAGE 8

8 LIST OF FIGURES Figure page 2 1 Heat evolution of Portland Cement ................................ ................................ ................... 19 2 2 Heat of Hydration Curve for Portland Cement ................................ ................................ 19 2 3 Scanning Electron Microscope ................................ ................................ ........................ 22 2 4 Beam interaction volume, showing different signal emis sion regions ............................ 23 2 5 Particle Size Distribution ................................ ................................ ................................ .. 26 2 6 ............................. 28 2 7 Initial microstructures ................................ ................................ ................................ ....... 33 4 1 Cement input Screen of the VCCTL ................................ ................................ ................. 43 4 2 Cement Phase Data Input Screen ................................ ................................ ...................... 45 4 3 Cement Particle Size Distribution ................................ ................................ ..................... 46 4 4 Microstructure Simulation Parameters ................................ ................................ ............. 47 4 5 Simple Isothermal Calorimeter ................................ ................................ ......................... 49 4 6 Admix Ampoule ................................ ................................ ................................ ................ 50 4 7 Tam Air Isothermal Calorimeter with sample and admix ampoule loaded ...................... 52 4 8 Hydration Simulation Input Parameters ................................ ................................ ............ 53 5 1 Saphir 550 Semi Automated Grinder Polisher ................................ ................................ .. 58 5 2 Optical Microscope ................................ ................................ ................................ ........... 60 5 3 Evaporative Carbon Coater ................................ ................................ ............................... 62 5 4 Cement Grain Close up ................................ ................................ ................................ ..... 63 5 5 BSE image as used for phase analysis ................................ ................................ .............. 64 5 6 Elemental Map for Sulfur ................................ ................................ ................................ 65 5 7 Maps required to distinguish Alite, Belite, Aluminate, Ferrite, and Gypsum .................. 65 5 8 Phase Classification ................................ ................................ ................................ .......... 66

PAGE 9

9 5 9 Image Classification Dialog ................................ ................................ .............................. 68 5 10 Statistical Analysis of Classification ................................ ................................ ................ 68 5 11 Segmented Image ................................ ................................ ................................ .............. 69 5 12 Segmented Image Prior to Thresholded Blur Operation ................................ ................... 70 5 13 Segmented Image after Blur ................................ ................................ ............................. 71 5 14 Fields Analyzed by UF and RJ Lee Group ................................ ................................ ....... 73 5 15 New false color image ................................ ................................ ................................ ...... 75 5 16 After processing, ready for classification ................................ ................................ ......... 76 5 17 Final Cement Image ................................ ................................ ................................ .......... 78 6 1 Continuous Measurements Display Page ................................ ................................ .......... 80 6 2 Fraction Solids Connected Curve ................................ ................................ ..................... 82 6 3 Periodic Measurement Display Page ................................ ................................ ................ 83 6 4 ................................ ................................ .. 88 6 5 Predicted Strength vs. time, w/c 0.4 ................................ ................................ ................. 88 6 6 Strength vs. Elastic Modulus, w/c 0.4 ................................ ................................ ............... 89 7 1 Preparation of Cylinders ................................ ................................ ................................ ... 92 7 2 Compressive Strength and Elastic Modulus Testing ................................ ........................ 92 7 3 Time of Set Test Apparatus ................................ ................................ .............................. 93 8 1 Compressive Strength vs. Time for different w/c ratios ................................ ................... 98 8 2 ................................ ......................... 99 8 3 Power vs. Time for different w/c ratios ................................ ................................ .......... 100 8 4 Energy vs. Time for different w/c ratios ................................ ................................ ......... 100 8 5 Young Modulus vs. Time, w/c ratio of 0.4 ................................ ................................ ..... 101 8 6 Compressive Strength vs. Time. w/c ratio of 0.4 ................................ ............................ 101 8 7 Young Modulus vs. Time, w/c ratio of 0.45 ................................ ................................ ... 102

PAGE 10

10 8 8 Compressive Strength vs. Time. w/c ratio of 0.45 ................................ .......................... 102 8 9 ................................ ................................ .. 103 8 10 Compressive Strength vs. Time. w/c ratio of 0.5 ................................ ............................ 103 8 11 ................................ ................................ 104 8 12 Compressive Strength vs. Time, w/c ratio of 0.55 ................................ .......................... 104 8 13 lorimetry Data ................................ ........... 105 8 14 .......................... 106 8 15 ................................ ... 106 8 16 odulus vs. Time, Larger Virtual Microstructure ................................ ........... 107 8 17 ................................ ............. 108 8 18 Fraction Solids Connected vs. Elapsed time ................................ ................................ ... 109 8 19 Measured Strength vs. Time for Different Admixtures ................................ .................. 110 8 20 Measu ................................ ... 110 8 21 ................................ ........................... 111 8 22 ................................ ......................... 112 8 23 ................................ ......................... 112 8 24 Strength vs Time, 4 oz/cwt WRDA 60 ................................ ................................ .......... 113 8 25 Strength vs. Time, 2 oz/cwt ADVA 600 ................................ ................................ ......... 113 8 26 Strength vs. Time, 4 oz/cwt ADVA 600 ................................ ................................ ......... 114 8 27 Power vs. Time ................................ ................................ ................................ ............... 114 8 28 Energy vs. Time ................................ ................................ ................................ .............. 115 8 29 Fraction Solids Connected vs. Time ................................ ................................ ............... 116 8 30 Penetration Resistance vs. Time, Different Admixtures and Dosages ........................... 116 8 31 Simulated vs. Measured Elastic Modulus ................................ ................................ ....... 117 8 32 Strength vs. Elastic Modulus with VCCTL Power Fit and ACI 318 .............................. 118

PAGE 11

11 8 33 Strength vs. Time, Different Empirical Predictions ................................ ........................ 119 8 34 Crushed Cylinder Fragments ................................ ................................ .......................... 121

PAGE 12

12 Abstract of Thesis Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Master of Science ANALYSIS OF THE SENSITIVITY OF A CONCRETE VIRTUAL MODEL TO DIFFERENT MATERIAL INPUT PARAMETERS By Benjamin E. Watts December 2013 Chair: Christopher C. Ferrar o Major: Civil Engineering The development, testing, and certification of new concrete mix designs is an expensive and time consuming aspect of the concrete industry. A software package, labeled the Virtual Concrete and Cement Testing Laboratory (VCCTL) h as been developed by the National Institute of Standards and Technology as a tool to predict the performance of concrete mixes quickly using computer simulation of the hydration behavior of concrete. This software requires specific data on the materials be ing simulated such as cement phase volume and surface area fraction, particle size distribution, gradation, density and other properties in order to accurately perform these predictions. A two phase testing program was implemented in conjunction with the Florida Department of Transportation to evaluate the effectiveness of the VCCTL for the prediction of concrete performance. The techniques required to characterize Portland cement were developed and implemented to provide input dat a for the VCCTL. The resulting virtual materials were simulated, and a testing program was performed on physical specimens to evaluate the accuracy of those simulations. Experimental values of compressive strength, elastic modulus, and time of set for diff erent mixtures were compared to the values predicted by the simulation of these mixtures within the software.

PAGE 13

13 CHAPTER 1 INTRODUCTION Background The development of a tool to predict the properties of Portland cement concrete has been the focus of many avenues of research and development. The complexity of the reactions that occur during cement hydration precluded the development of useful computati onal models for the process un til the late 1980 s due to the limitations of available computational resources Over the last 12 years, software known as the Virtual Cement and Concrete Testing Laboratory (VCCTL) has been available for commercial use from Th e National Institute of Standards and Technology (NIST) This software incorporates microstructural modeling of Portland cement hydration, and allows for the prediction of different properties of the hydrated product. The efficacy of the model relies on th e proper characterization of the materials being simulated. While the potential usefulness of this tool is substantial, its accuracy, particularly with regards to materials endemic to the state of Florida and the simulation of the properties of concrete sp ecifically (as opposed to mortar or grout) have yet to be systematically evaluated. Model Function The VCCTL uses data from real materials to create a virtual concrete from which different material properties can be obtained via virtual testing. The model first creates a three dimensional representation of a Portland cement suspension, as would exist at the initial moment of mixing the cement and water. This composition of this representation is drawn directly from the data supplied for the cement being mo deled. In a process that mimics the actual hydration of Portland cement, t he model virtually hydrates this three dimensional microstructure using a specific set of rules drawn from the observed hydration kinetics and thermodynamics of Portland cement. As t he virtual microstructure is hydrated, parameters such as heat released, total

PAGE 14

14 porosity, degree of hydration, and many others are calculated in real time. When hydration is complete, a finite element calculation run on the microstructure provides the elast ic modulus of the paste. This elastic modulus can then be used in combination with the elastic properties of the coarse and fine aggregate to predict the modulus of the concrete itself. Finally, compressive strength is predicted from this modulus using a s imple empirical relationship. Research Requirements Evaluating the accuracy of the VCCTL requires the comparison of predicted properties for a given concrete from the model with actual properties of physical specimens. Elastic modulus, compressive strength and time of set are easily measureable predictive outputs of the model, and knowledge of the actual values of these properties for a given concrete is essential. It is also critical that proper characterization of the materials being simulated is perform ed. The existing methods and procedures required for this characterization must implemented and refined. Research Objectives The primary objective of this research is to determine the degree of accuracy with which the VCCTL is capable of predicting the var ious properties of Portland cement concrete. The specific objectives are as follows: Determine the accuracy with which the VCCTL predicts the e lastic m odulus of Portland cement concrete Determine the accuracy wit h which the VCCTL predicts the c ompressive s trength of Portland cement concrete Determine the accuracy with which the VCCTL predicts the set time of Portland cement concrete on Set Time through isothermal calorimet ry.

PAGE 15

15 Evaluate the influence of different material data on the degree of accuracy with which the above properties are predicted A secondary objective of this research is to examine the possibility of expediting the processes required to characterize the different materials being simulated. Hypothesis The Virtual Cement and Concrete Testing Laboratory predicts the properties of concrete by modeling the hydration of Portland cement microstructure. The accuracy with which this simulation takes place is highl y dependent upon the accuracy with which the materials being simulated are characterized. The processes by which these materials are characterized may also be made faster and more accurate through the development of new techniques and the utilization of mo dern instrumentation. Significance of Research of concretes containing materials endemic to the state of Florida. If the VCCTL can be proven to be an accurate tool for t he prediction of the properties of Portland cement concrete, it could dramatically reduce the time and labor required for the design and verification of new concrete mixes. The VCCTL may also be a valuable exploratory tool to examine the impact of a change in mix design on the performance of an already established mix. Outline of Thesis Chapter 2 of this thesis comprises a brief summary of the basics of Portland cement composition and hydration mechanics as well as a review of literature relevant to the dev elopment and history of the VCCTL. Chapter 3 discusses the approach taken while performing this research. Chapter 4 is a discussion of the material data required as inputs for the VCCTL, as well as the methods by

PAGE 16

16 which they are acquired for all but one of these data. One of the inputs for the model requires a specialized technique, the development and implementation of which comprised a large portion of this project. This te chnique is detailed in Chapter 5 The data provided by simulations run within the V CCTL is discussed in Chapter 6, while Chapter 7 details the physical testing program. Chapter 8 comprises a comparison between the results of the physical testing program and the results of simulations run within the model. The conclusions drawn from this compar ison are summarized in Chapter 9 and recommendations for future re search are outlined in Chapter 10

PAGE 17

17 CHAPTER 2 LITERATURE REVIEW Portland Cement Hydration When mixed with water, Portland cement undergoes a complex set of reactions that res ult in a transformation of a colloidal suspension of cement particles in water to an interconnected matrix of hydration products. This rate and resulting products of this reaction are governed largely by the relative concentrations of the four major consti tuents of Portland cement: a lite (C 3 S), b elite (C 2 S), a luminate (C 3 A), and f errite (C 4 AF). A fifth mineral component, g ypsum (CaSO 4 2 H 2 O), plays an important role in the early stages of the hydration reaction (Mindness & Young, 1981) Aluminate is the most readily soluble of the compounds present in Portland cement, to the point that unchecked, it will result in an immediate stiffening of the cement paste, known as t by reacting with the aluminate to form insoluble compounds (Neville, 2011) Gypsum is not a result of the reactions that produce Portland cement, and is usually added to the clinker before grinding. Ideally the amount of gyps um added is just enough to leave very little aluminate available for direct hydration. In addition to gypsum, other calcium sulfate minerals such as anhydrite (CaSO 4 ) or hemihydrate (CaSO 4 0.5 H 2 O) can be added to have a similar effect. As with gypsum, t he addition of these sulfates must be carefully controlled. The majority of the strength of hydrated Portland cement comes from the hydration o f the Calcium Silicate phases (a lite and b elite ). These phases react with water to produce Calcium Silicate Hydrate (CSH) and Calcium Hydroxide (CH). While both alite and belite react with water to produce CSH, the stoichiometry and rates of their respective reactions differ. The reaction of alite with water occurs more quickly than that of belite, and is largely responsible for

PAGE 18

18 strength development at early ages (up to 28 days). Alite also produces more of the water soluble CH relative to CSH. Belite reacts more slowly, producing less CH relative to CS H and contributing primarily to strength development after 28 days. kilning of the raw materials required to produce Portland cement clinker, the ferrite that is form ed becomes liquid and acts as a flux, lowering the temperatures required for the reactions that produce alite and belite. Gypsum does react with ferrite, and the products of that reaction may have an accelerating effect on the hydration of the silicates. Cement Heat of Hydration The reactions that occur as cement hydrates are exothermic, that is they produce heat as the reaction occurs. Heat evolution during cement hydration is neither linear nor continuous, but does generally follow a specific pattern. A typical heat evolution curve is shown in Figure 2 1. The time elapsed in this Figure 2 1 refers to the amount of time that has passed since the mixing of cement and water. The initial spike in Figure 2 1 occurs almost immediately, and corresponds to the di ssolution of C 3 A. The rapid hydration of C 3 A is followed by a dormant period during which the paste is workable. This is typically the time during which placement of concrete would occur. The dormant period ends and heat is released more gradually as C 3 S a nd C 2 S hydrate and paste. The rate of this reaction peaks at approximately 10 hours (Neville, 2011) after which it tapers off. There is a third peak that is visible at roughly 13 hours in Figure 2 1; this peak corresponds to a renewed reaction of C 3 A following the exhaustion of gypsum (Neville, 2011) The reaction slows slowly after this third peak, as diffusion throug h the crowded microstructure begins to limit the rate of reaction. The total heat evolved can be taking the area under the power curve shown in Figure 2 1 to obtain a heat of hydration curve, shown in Figure 2 2.

PAGE 19

19 Figure 2 1 Heat evolution of Portland C emen t Figure 2 2 Heat of Hydration Curve for Portland Cemen t Admixtures Chemical admixtures, while not essential for the production of concrete, are used almost ubiquitously to alter the behavior of fresh and hardened concrete. The most common types of admixtures are those which alter the viscosity of the fresh concrete, and allow for the reduction 0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00 8.00 0.00 5.00 10.00 15.00 20.00 25.00 30.00 35.00 40.00 45.00 Power (mw/g) Time (hours) 0.00 50.00 100.00 150.00 200.00 250.00 300.00 350.00 400.00 0.00 20.00 40.00 60.00 80.00 100.00 120.00 140.00 160.00 180.00 Energy (J/g) Time (hours)

PAGE 20

20 of water to achieve a higher strength without the sacrifice of favorable plastic properties. These admixtures are referred to as water reducing admixtures, an d operate via a fairly simple mechanism. Water reducing admixtures can be broken down very loosely into two categories; normal water reducing admixtures and high range water reducing admixtures. The mechanism by which these admixtures op erate is fundamenta lly the same; however their composition and the degree to which their effects are manifested in concrete vary substantially. Normal water reducing admixtures, classified as Types A or D, typically consist of either lignosulfate or hydrocarboxylic acids. T hese compounds act by adsorbing onto the cement particles and surrounding them with an envelope of negative charge. This results in mutual repulsion between the particles, causing them to disperse more evenly (Mindness & Young, 1981 ) This manifests on a larger scale as a reduction in viscosity, due to both the mutual repulsion of the particles and the availability of water for lubrication that would normally be adsorbed onto particle surfaces. High range water reducing admixtures (also referred to as superplasticizers), operate in much the same way as normal water reducers, but to a greater extent. Chemically, high range water reducers are typically composed of synthesized long chain organic pol ymers. These compounds work more efficiently than normal water reducing agents, exhibiting a stronger negative charge when adsorbed onto the surface of cement particles. The corresponding reduction in viscosity is also more substantial (Edmeades & Hewlett, 1998) Though normal and high range water reducers differ in composition and in the degree to which their effects manifest, the similarity of the mechanisms by which they operate results in several common effects. Due to the m ore uniform dispersion of the cement particles, more

PAGE 21

21 surface area is available for hydration to occur, which can result in higher strengths at early ages (Neville, 2011) They can also have a retarding effect on the hydration r eaction because the surrounding of the cement particles with admixture compounds temporarily limits the surface area available for reaction to occur. Normal water reducers that exhibit this effect are classified as Type D, while those that do not are class ified as Type A. Strength of Concrete The most frequently used industry metric for the evaluation of a concrete mix design is strength. Strength of concrete is most commonly taken to mean uniaxial compressive strength, which is typically obtained by crushi ng cast cylinders. The strength of the concrete is influenced by a number of factors, including the relative concentrations of the different phases of Portland cement, however the water to cement ratio (w/c) is typically regarded to have greatest effect on strength, all else being equal. This occurs primarily due to two factors. The more important of these is a reduction of the gel/space ratio the ratio of the volume of hydrated to cement paste to the sum of the volumes of the hyd rated cement and capillary pores (Neville, 2011) This can more simply be thought of as the density of the cement paste, as it follows that the more water is present relative to the much heavier cement, the lighter the resulti ng paste will be. Gel/space ratio and water/cement ratio are inversely related, that is the lower the water to cement ratio, the higher the gel to space ratio. As water/cement ratio decreases and gel/space ratio increases, strength also increases. The ot her main mechanism by which water/cement ratio influences the strength of concrete is related to the relationship between the coarse aggregate and the hydrated cement paste. As the water to cement ratio increases, bleeding (separation of the water from the paste) begins to occur around the aggregate particles. This bleeding results in cracks around the aggregate particles that are prone to failure under stress (Maso, 1996) Th e extent to which this

PAGE 22

22 phenomenon occurs depends on t he amount of water in the paste, with lower water/cement ratios being less affected. Scanning Electron Microscopy Figure 2 3 Scanning Electron Microscope (Goldstein, et al., 2007) Scanning electron microscopy (SEM) is a mic roanalysis technique that allows for the imaging of features several orders of magnitude smaller than those that can be imaged using conventional optical microscopy. The fundamental operation of a SEM relies on the interaction of a focused electron beam wi th the surface of the specimen being imaged. This electron beam is created using a thin tungsten filament, which when subjected to heat and a very strong accelerating voltage, results in the emission of electrons. The electrons are focused using a series o f lenses and apertures to a spot on the surface of the specimen. A schematic of a typical instrument is shown in Figure 2 3.

PAGE 23

23 The electrons of beam penetrate into the surface of the sample and collide into the atoms present. The region in which this occurs is referred to as the i nteraction volume, a diagram of which is shown in Figure 2 4 Figure 2 4 Beam interaction volume, showing different signal emission regions (Wittke, 2008) The interaction of the electrons with the atoms in the specimen result in the emission of several different types of signal that can be collected and used to create an image. There are two primary types of emitted electrons that are used to create an image backscattered electrons and secondary electron s. Backscatter electrons (BSE) are emitted from the sample when electrons from the beam undergo elastic collisions with the atoms in the sample and are scattered back out of the surface of the sample. These electrons have high kinetic energy and are detect ed by an annular detector that surrounds the aperture from which the beam originates. The other type of electrons that are emitted from the sample are secondary electrons (SE) which undergo inelastic scattering events with the atoms in the sample and lose much of their kinetic energy in the

PAGE 24

24 process. These electrons that make it back to the surface of the sample are attracted toward a positively charged detector and used to create signal. Characteristic X ray photons are also emitted from the interaction vo lume as a result of inelastic scattering events. These photons have characteristic levels of energy depending upon the element with which they interact. These photons allow for the elemental analysis of a specimen, and this form of imaging is also referred to as energy dispersive X ray spectroscopy or EDS. The actual image in a SEM is produced by collecting the signal of interest from the spot, and moving the spot rapidly in a grid pattern. The beam moves very rapidly through this grid pattern, and the diff erent signal intensities at each spot (which corresponds to a pixel) produce contrast in the image. The contrast mechanisms of the different imaging modes are the result of the properties of the sample. BSEs are produced as a result of elastic scattering e vents, which occur more frequently in high atomic number samples. This results in an image where the brightest areas have the highest average atomic number. Secondary electrons are emitted more from topographic irregularities on the surface of the sample; this imaging mode shows relief or scratches in the sample surface very well. X ray photons, also the result of inelastic scattering, are emitted preferentially by lower atomic number elements, and by topographic deviations due to the s hape of the interacti on volume. Though SEM allows for higher magnifications and different imaging mode s than conventional optical microscopy, certain aspects of the imaging process limit the types of samples that can be studied The whole SEM is under vacuum, as gas would ion ize in the electron beam if present. This precludes the imaging of samples containing water or any other compounds that would boil away at very low pressures. The sample must also either be

PAGE 25

25 conductive or be made conductive, otherwise electrons will charge the surface of the specimen, resulting in bright artifacts in the image. For elemental analysis using EDS, the sample must be reasonably flat, as topographic variation can influence the degree to which X ray s are emitted from different parts of the sample. Many of these restrictions are dealt with careful preparation of specime ns. The procedures used for this research are detailed in the beginning of Chapter 4. Computer Modeling Since the invention of the transistor and the subsequent exponential growth of computational processing power, efforts have been made to si mulate complex systems through t he use of computer models. A model can be quite simply defined as a representation of a system upon which operations can be performed to examine the response of th e system to specific stimuli. The engineering applications of computer modeling typically extend to numerical or stochastic simulations, the former describing a system governed by equations that cannot be solved analytically, and the latter describing a sy stem where the occurrence of events is probabilistic in nature. Both types of simulations have applications in civil engineering, with numerical modeling of structural systems using the finite element method being extremely common, while stochastic simulat ions are frequently applied to the simulation of different weather phenomena. The finite element method is a technique used to numerically simulate the overall behavior of geometrically complex objects by discretizing them into many interconnected element s each of which can be described with its own set of equilibrium equ ations. Given a certain set of boundary co nditions which act on the geometry, the system of equations from the individual elements can be solved to determine the influence of the boundary conditions on the entire system (Zeinkiewicz, Taylor, & Zhu, 2005) A common example application of this method is the analysis of the deflection of a truss structure with all pinned connections under a

PAGE 26

26 specific loading condit ion. As long as the structure is in static equilibrium, the forces in each element (truss) the structure is discretized into must also be in equilibrium. Because the relationship between force and displacement for each element is known, the system of equat ions for all elements can be solved to find the resulting displacement of the structure as a whole. Stochastic modeling is used to simulate the behavior of systems that are non deterministic, as in they e xhibit behavior that contains an element of randomne ss. These s ystems can instead be described by the probabilities that events within the system will o ccur. For example, the range of particle sizes that make up Portland cement is highly variable. If a number of particles for a particular cement are analyzed, the probability that a particle will be of a particular size (the event) can be obtained by dividing the number of times a particle that fell within a given size range was observed by the number of particles that were measured. Figure 2 5 Par ticle Size Distribution If all particles are measured and classified into different size ranges, and the probability that a particle is a given size is charted against the size range, the result is a chart that looks like Figure 2 5. 0 1 2 3 4 5 6 7 8 9 0.1 1 10 100 1000 % Measured Particles Particle Diameter (m)

PAGE 27

27 This can also be descr ibed as the probability density function for the size of Portland cement particles. This information can be used to create a set of virtual cement particles with the same distribution of sizes. Cementitious Simulation Background The fundamental goal of mod eling cementitious systems is the prediction of the structural and durability properties of the material being simulated . Properties of the hardened cementitious microstructure that are desirable include elastic properties such as elasticity and ultimate strength, or properties such as porosity or permeability which are more related to the durability of the cement paste. The creation of structures made from concrete relies on the knowledge of these characteristics of the material. The accurate determinatio n of these properties via computer simulation depends on the creation of a realistic virtual microstructure from which they can be measured. Because the properties of the microstructure are dependent upon the composition and hydration conditions of the cem ent from which it is created, there must also be a method by which to simulate the development of the microstructure over time. History The origins of cement microstructural modeling come from research done in the 1970 s on the structure of amorphous semic onductors (Garboczi E. J., 2000) Initial attempts to calculate structural properties using approximate analytical solutions were only marginally successful. A model was then constructed consisting of several hundred randoml y linked atoms. Operations were performed on this model to calculate properties, and the resulting properties were compared to experimental results. Early models like this represented the first attempts to model amorphous materials computationally at the a tomic level.

PAGE 28

28 The first attempt to model concrete computationally came with the publication of a paper by Wittman, Roelfstra, and Sadouki in 1984 on the numerical simulation of the structure and properties of concrete in two dimensions (Wittmann, Roelfstra, & Sadouki, 1984 1985) The two dimensional silhouettes of aggregates were first characterized by transforming the contour of an aggregate particle section into polar coordinates. The radius of the particle as a function of t he angle theta about the y axis could then be plotted, and the resulting frequency distribution obtained. The frequency distributions of several particles of a given aggregate normalized for was then used to generate aggregate sections computationally. Figure 2 6 (Wittmann, Roelfstra, & Sadouki, 1984 1985)

PAGE 29

29 The next step was the computer generation of a two dimensional section of a composite concrete matrix using this morphological law in combination with a measured aggregate size distribution (Figure 2 6). This research goes on to acknowledge the possibility of generating a three dimensional concrete structure using the same principles, but rules out the analysis of this three dimensional structure due to the limitations of computational resources at the time. An alternative approach was used to represent the three dimensional composite structure using two dimensional images. To obtain a two dimensional image representative of the three dimensional structure, the size distributions of aggregate particles for a number of arbitrarily chosen planes from the three dimensional structure were averaged. This average distribution was then used to create a representative two dimensional composite structure. The two dimensional structure was represented by a finite element mesh, which combined with an assumed value for t he modulus of the cement paste allowed for the computation of the elastic modulus of the composite as a function of aggregate modulus. Different meshing techniques allowed for the computation of the effective diffusion coefficient of the composite structur e. Finally, it was projected further work in which a mesh containing more detailed material properties could be used to predict more complex behavior, such as creep, shrinkage, and non linear stress strain behavior. In 1986 the details of a three dimensional hydration model for C 3 S were published (Hamlin N. Jennings, 1986) The paper begins with a brief overview of the mathematical models for hydration at the time, which typically utilized differential equations that in corporated the different kinetic and chemical mechanism s by which the reaction was understood to occur. The equations used in this method were observed to have questionable physical significance. The

PAGE 30

30 creation of a model that instead follows rules based on measureable characteristics of the system being simulated was then proposed as an alternative. The primary focus of this model was not to accurately simulate the hydration behavior of C 3 S, but instead to provide a tool that connects the probable mechanisms by which the reaction occurs with the measurable behavior of the system. Such a tool can be useful to evaluate the validity of a proposed mechanism by simulating its effect on the behavior of the system. The model simulated C 3 S particles as spheres, with the size distribution, number, and initial packing type of the spheres input by the user. Other user controlled inputs included the density of different hydration products, rules governing the distribution of hydration products and the rate controlling rea ction step at each stage. Simulations within the model began by distributing the input number of particles with the specified size distribution randomly. Particles were hydrated incrementally in sequence. The diameter of the smallest particle was reduced by an amount governed by the input rate controlling step as well as information about the rate determining step. The space left by the reduction in diameter was filled with hydration product surrounding particle, and any remaining hydration product was add ed to the diameter of the particle. This process is repeated for each particle in sequence. One hydration cycle was complete when all particles had undergone this process. The simulation continued until either all anhydrous phases were consumed, the thickn ess of the hydrated product layer reached a user specified value, or a specific number of cycles were completed. The resulting microstructure from simulations with this model was compared to SEM micrographs of the hydrated C 3 S grains. A promising degree o f resemblance was observed. The future objectives of this research envisioned the use of the model for the evaluation of different

PAGE 31

31 mechanical properties of the composite structure, and predicted the model would enable the research into proposed reaction me chanisms. The incorporation of the other phases of Portland cement was also envisioned. This model was used as a tool for the simulation of cement hydration products for several years, including work to apply a random walk algorithm for continuum models ( including the Jennings and Johnson model) in order to compute the electrical and diffusive transport properties of the microstructure (Garboczi E. J., 2000) It was during this work that the idea of digitizing the microstruc ture was experimented with due to the difficulty of computations on a continuum based microstructure model, as well as the fact that the available computing power had just reached the point at which a model of sufficient size was feasible. The following de scribes what exactly a digitized microstructure is. A digital microstructure is essentially a three dimensional image consisting of voxels (3D pixels). Each voxel represents a discrete volume of material with specific properties. A virtual microstructure i s created out of the voxels based on the physical and chemical characteristics of the cement being modeled. Once the input parameters have been entered into the computer for simulation, the virtual microstructure undergoes the simulated hydration process, with each voxel acting as an independent agent. Voxels follow specific rules for dissolution, diffusion, and reaction based on their phase and the known thermodynamic and kinetic behavior of cement hydration. The creation of this digital microstructure led to the realization that nearly any fi nite element or finite difference algorithm could now be applied to measure the properties of the microstructure. The only remaining steps were to establish the applications of percolation theory and composite material theory to the digital microstructure.

PAGE 32

32 Percolation generally refers to the slow trickle of a fluid through small but interconnected porosity. Percolation theory is based on the idea of connectedness in a random process. For the purposes of microstructural modeling of cement this applies more specifically to the connectedness of random phases in a multiphase environment (Garboczi & Bentz, 1998) The applications of this concept to the measureable properties of a hydrating mic rostructure are twofold. The degree of percolation of pore space limits the available paths for which diffusion can occur, and the percolation of solids results in set, and as hydration progresses limits reduces the percolation of the pore space. The appli cation of percolation theory to a digital lattice was established through work on the conductivity of a plane containing random holes (Garboczi, Thorpe, & Day, 1991) and algorithm was developed to compute the linear elastic properties of a heterogeneous materials from a digital lattice a few years later (Garboczi & Day, 1995) The combination of different digital lattice based methods for the measurement of cement microstructure let to the even tual development of a three dimensional cement hydration model known as Cemhyd3d in 1997, the model that in an updated form underpins the VCCTL. The VCCTL was introduced by the National Institute for the Standards and Technology (NIST) in 2001 and is inten ded as a unified solution for modeling concrete at multiple length scales. Concrete can be described as a multi scale material, in that the properties of the paste microstructure which exist at a micro (10 6 ) scale influence the properties of the paste at a millimeter scale which in turn influences the properties of the concrete at a scale level. The intention of the VCCTL is to be a start to finish modeling solution for concretes and mortars, requiring only data on the materials being simulated to provid e accurate measurements of mechanical and transport properties.

PAGE 33

33 CEMHYD3D Cemhyd3d, the hydration model used by the VCCTL, has been available for public use since 1997. Publications of note from this period include a quantitative comparison of the initial a nd hydrated microstructures generated by Cemhyd3d (Bentz D. P., 2005) and an investigation of the influence of ground limestone filler on cement hydration (Bentz D. P., 2006) Figure 2 7 Initial microstructures of (a) real cement, (b) model with spherical particles and (c) model with real particle shapes (Bentz D. P., 2005) The intention of the comparison of simulated and real cement microstructures was to prove the validity of CEMHY3D for the modeling of Portland cement hydration. Microstructures were compared visually and also quantitatively using a normalized two point correlation

PAGE 34

34 function. The initial microstructure of a cement was compared to model simu lations using spherical particles as well as real particle shapes. The visual comparison of the different microstructures can be seen in Figure 2 7 Image sets such as the one shown in Figure 2 7 were also compared for hydrated model and real microstructure s. Conclusions made from this research were that CEMHYD3D agrees very closely with the real systems, both in the visual comparison of model and real microstructures, and the comparison of two point correlation functions for model and real microstructures. It was also observed that the model microstructures with real particle shapes agree better with the two point correlation functions of the real cement, and that larger model system sizes also improve the agreement between these functions. The investigation of the influence of finely ground limestone on the hydration of cement was triggered by a change in the ASTM C 150 standard specification for Portland cement to allow for the inclusion of up to 5% of ground limestone, and the claim that this does not infl uence the performance of Portland cement. CEMHYD3D was modified for this study to account for the hydration behavior of cements containing ground limestone. The main modifications to the model took into account the potential for the additional surface area provided by the limestone to provide more nucleation sites for the growth of hydration products, as well as the potential for a chemical reaction to occur between the cement and limestone. The results of the hydration model were validated against experime ntal measurements of degree of hydration via non evaporable water content. This study concluded based on good agreement of model and experimental data that CEMHYD3D successfully adapted to the chemical and surface area effects of ground limestone.

PAGE 35

35 Existing Research using the VCCTL The VCCTL has been available for commercial and research use for over a decade, and in that time has been continually developed and improved. Because of the continual improvements made to the model during this time period, prefere nce was given to more recent research into the applications of the VCCTL. One of the simplest applications of the VCCTL is as an exploratory tool to investigate the effects of changes to cement chemistry. The ability of the model to consider changes not on ly to cement composition but also fineness (via particle size distributions) allows for optimization of the cement being produced with a minimum of additional labor. Work done by Siam Cement Group (SGC) (Sahachaiyunta, 2012) in vestigated the influence of sulfate content and fineness on mortar strength. Twenty five combinations of fineness and sulfate content were simulated at 3, 7 and 28 day ages. The strength of the mortars was calculated for all twenty five combinations at dif ferent ages and used to evaluate the performance of each permutation. It was concluded from results of plant trials that the VCCTL could be an effective tool for the optimization of different cement production parameters, especially when considering more complex blended cements or the implementation of high risk (economically) production strategies. While the use of the VCCTL as an optimization tool in a cement plant is a valuable application of its functionality, the primary focus of this research is the prediction of the properties of concrete. Some work in this direction has been done with Mapei Industries in conjunction with the University of Padua (Valentini, et al., 2013) The accuracy of both the hydration simulation and m odulus calculation of the VCCTL was evaluated for two cements. An in situ X ray diffraction technique was used to measure the changing volume fractions of different cement phases and hydration products over time. For elastic modulus, the predictions of

PAGE 36

36 the model were compared to measurements taken from mortar cubes. Additionally, a power law equation was developed to empirically predict the strength of mortars from elastic modulus. This study found that the VCCTL simulates the hydration behavior of the chi ef hydration products, CSH and CH, quite well, and that it also simulated the diminishing quantity of C 3 S with a reasonable degree of accuracy. There were substantial differences between the measured and predicted volume fractions of ettringite, however it was concluded that the low concentration of ettringite in the system rendered the inaccuracy largely irrelevant for the prediction of elastic and mechanical properties. The study found good agreement between prediction and measurement of elastic modulus a nd the empirical relationship developed for the prediction of strength from modulus also correlated well with measurements. Of note is the strong caution against the use of the empirical equation for strength developed in this study for mortars in general, as the properties of the input materials may vary substantially. Another current focus of research with the VCCTL is the characterization of admixtures. The VCCTL has the capacity to handle admixtures, provided the behavior of the admixture is known with respect to cement phase surface deactivation time. This information is not commonly available for admixtures however, and a method by which it can be obtained is currently in development. The research examines the behavior of admixtures by filtering cement and pure phase pastes via gel permeation chromatography of the resulting liquid. From this the adsorbed amount of superplasticizer can be determined based on the specific surface area of the cement particles from BET gas adsorption. This then allows for the determination of the phase surface deactivation time for input into the VCCTL (Russo, 2012) This technique is still in development, and has yet to be verif ied.

PAGE 37

37 There is no research currently published that investigates the accuracy of the VCCTL when concrete is being modeled, as opposed to mortar. These studies do support the accuracy of the VCCTL in the prediction of the properties of the cement paste as we ll as mortars made with that paste, however the methods by which the VCCTL incorporates coarse aggregate have yet to be systematically validated. Current Limitations of the VCCTL Though the VCCTL is the product of years of research into the modeling of cem entitious systems, there are still some aspects of concrete for which it is not as robust. One of these areas granulated blast furnace slag. These are products that are used in almost all concrete mixture designs today. The Florida Department of Transportation, for instance, has a minimum fly ash requirement for all structural concrete mixture designs. The VCCTL has some ability to handle these materials, however it has not been very well validated and is missing a lot of details as far at the thermodynamics and specifics of reactions are concerned (Garboczi, Bullard, & Martys, 2010) The VCCTL also has limited support for admixtures, but i t is implemented in a way that makes them difficult to use currently. The model requires that the specific amount of time that the surface of a phase is deactivated by the admixture is known. This information is difficult to obtain. Efforts are currently u nderway to develop a process by which admixtures can be characterized to obtain this information. It may also be possible to obtain this information from the manufacturers of the admixtures themselves. Durability is one of the most important aspects of con crete, both for the purposes of sustainability and economy. Any predictive model for the performance of concrete would likely benefit from some sort of durability prediction and degradation modeling. Unfortunately, the

PAGE 38

38 VCCTL does not currently provide any direct functionality for the prediction of durability in concrete. Its hydration model does provide several outputs that can be used for additional modeling to simulate the durability of concrete, however no direct outputs exist. Other Cementitious Modelin g Packages The prediction of concrete performance is a problem for which a solution has great economic potential. As a result, several commercial modeling solutions for concrete exist. None of these commercial solutions provide results that are as fundamen tal as those simulated by the VCCTL, but their more specific focus has enabled them to be commercially successful. One of the most well recognized modeling solutions for cementitious systems is STADIUM, which is a model that predicts the service life of co ncrete mixtures based on material testing. A mixture design that is being considered for use is tested for a number of properties, such as porosity, which are then compared to large databases of similar mix designs under different exposure conditions. The model is capable of predicting service life for many different mixture designs in many different environmental conditions, and has seen use by organizations such as the US Navy. Though the service life predictions of STADIUM have reached a degree of commer cial acceptance, the models and databases from which it makes these predictions are proprietary. The exact details of the function of STADIUM are not documented. Some work has been done at Mines ParisTech on a numerical cracking model to simulate the effec ts of Alkali Silicate Reaction (ASR) (Peyrot, 2006) The model uses a finite element approach to simulate the growth of reactive aggregates bound in a matrix, and the resulting bulk swelling. This model incorporates methods for simulating crack development and propagation in heterogeneous materials. Though this direction of research is promising, this model is still in the research stages, and has not seen any commercial use.

PAGE 39

39 Another concrete material model that has seen a degre e of commercial acceptance is TNO DIANA, a finite element model for reinforced concrete structures using realistic material properties for concrete and steel. The chief distinguishing feature of TNO DIANA is the incorporation of failure analysis for reinfo rced concrete structures, and the detail with which it models the constituents of reinforced concrete. The software actually incorporates concrete cracking and crushing models for the purposes of modeling reinforced concrete failure. On the opposite end o f the size spectrum, the Massachusetts Institute of Technology (MIT) Concrete Sustainability Hub is currently working in cooperation with the Portland Cement Association (PCA) to understand and model the atomic structure of Portland cement concrete. The go als of this Concrete Sustainability Hub are to bring about breakthroughs in concrete science and transfer those advancements to the concrete industry. One of their key goals is the reduction of Portland Cement Productions contribution to CO 2 emissions. Currently a number of advances in the understanding of the nano scale atomic structure of certain phases in cement have been realized, but a practical link between any of these advances and industry applications remains to be seen. The speciali zed nature of these models differentiate s them from the VCCTL, which is intended to be a full spectrum solution for the modeling of concrete. The further development of the VCCTL to incorporate features such as fracture analysis and service life prediction is ongoing.

PAGE 40

40 CHAPTER 3 RESEARCH APPROACH The VCCTL requires detailed data on the material being simulated within the model to research, are obtained via th e thorough characterization of the materials in the laboratory being considered for this study. Some of this material data is obtained through standard tests, however the implementation of more involved characterization techniques is necessary for the acqu isition of certain material properties. The development and improvement of techniques for the characterization of materials was an initial goal and large part of the work performed for this research. The testing program for the development of material inpu ts utilized following material analysis techniques Scanning electron microscopy and Energy Dispersive X ray Spectroscopy microanalysis Isothermal Conduction Calorimetry Laser Particle Size Distribution Analysis X ray Powder Diffraction (XRD) analysis Th ese characterization techniques provided the following inputs for use with the VCCTL. Volume and Surface Area Fractions for the different primary Portland C ement phases Heat of Hydration of Portland cement Particle Size Distribution of Portland Cement Mass Fractions of Sulfate Phases in Portland Cement for different prop erties of concrete. A study to evaluate the predictions of Elastic Modulus, Compressive strength, and T ime of Set for concrete was constructed to gauge the accuracy of the VCCTL. Secondary aspects of this study included a sensitivity analysis of the VCCTL to certain material inputs, as well as an experimental technique to calibrate the model to the behavior of different admixtures. A physical testing program was implemented to provide data on simple

PAGE 41

41 concretes against which to compare the results of the model. The results of the following tests on fresh and hardened concrete were performed. Air Content (ASTM C 173) Slump (ASTM C 143) Unit Weight (ASTM C 138) Temperature (ASTM C 1064) Time of Set (ASTM C 403) Compressive Strength (ASTM C39) Compressive Elastic Modulus (ASTM C469) Of these tests, time of s et, e lastic m odulus and c ompressive strength were compar ed to predictions made by t he model. Other tests, such as s lump, u nit w eight, t emperature, and a ir c ontent were performed as part of standard mixing procedure for quality control purposes.

PAGE 42

42 CHAPTER 4 VCCTL MATERIAL INPUTS Overview The most important part of using the VCCTL is the proper characterization of the materials being simulated. The model takes a number of different types of input data, some of which are essential to running a simulation, and others which simply serve to enhance the accuracy of si mulations, or alter the way in which the virtual microstructure is generated or hydrated. The inputs that are required for any hydration simulation include the following Cement phase volume and surface area fractions Cement phase relative spatial distribut ions Cement particle size distribution If running simulations including aggregates (mortar or concrete), the following inputs are also required. Aggregate bulk and shear moduli Aggregate gradation Aggregate specific gravity There are other inputs that are not required by the VCCTL to run simulations; however it is possible that the presence of these inputs may improve accuracy. For the purposes of this Cement heat of hydration curve Mass fraction data of calcium sulfate phases present in cement Particle size distribution of sulfate phases Aggregate shape data Cement particle shape data Larger simulated microstructure dimensions Cement particle dispersion within microst ructure Other inp uts that the VCCTL will accept includ e data on the composition and particle size distributions of pozzolonic materials such as fly ash or slag, particle size distribution data for

PAGE 43

43 the different sulfate phases present in cement, or the che mical surface deactivation behavior of admixtures. While methods for obtaining theses inputs exist, they are either difficult to obtain, not within the scope of this research project, or acceptable substitutes already exist within the rial library. Inputs for Portland Cement Figure 4 1 Cement input Screen of the VCCTL The inputs required for Portland cement can be divided into two categories : inputs that are used to build the digital microstructure, and inputs that are used during th e hydration of the

PAGE 44

44 microstructure. Many of these inputs can be obtained fairly simply via standardized tests and/or specialized equipment, while others require more complex procedures. Cement material inputs are entered into the VCCTL through the Lab Mater ials Cements page, shown in Figure 4 1. Details of the specific inputs for Portland cement are as follows. Microstructural Inputs The following inputs are used by the VCCTL to create a digital microstructure that is representative of the cement from whi ch they are obtained. Gypsum Phase Mass Fractions ASTM C 1635 Standard Test Method for Determination of the Proportion of Phases in Portland cement and Portland ce ment Clinker Using X ray Powder Diffraction Analysis is used to for direct determination of the mass fractions of the different phases present in Portland cement. This test measures the crystalline nature of the phases present in Portland cement. By measuring the angle of the fo rmation of a diffraction pattern relative to the incident X ray s, the nature of the material present can be determined (Waseda, Matsubara, & Shinoda, 2011) Within the VCCTL, this test is primarily used to obtain the mass fract ions of the different calcium sulfate phases, the input fields for which are visible in Figure 4 1. Because this test provides mass fraction data for all the major chemical constituents of Portland cement, it also plays an essential role in the validation of the cement phase volume fraction data. Cement Phase Volume and Surface Area Fractions, Two Point Correlation Functions The most important data required by the VCCTL on the cement being simulated is the volume and surface area fractions of the four major cement phases: alite, belite, aluminate and ferrite. While volume fraction can be obtained via X ray powder diffraction, acquisition of surface area fraction data is much more complex, involving scanning electron microscopy and multispectral image analysi s. This process results in a false color segmented image of a cross

PAGE 45

45 section of cement particles. The complete process of cement phase volume and surface area acquisition is detailed in Chapter 4. Figure 4 2 Cement Phase Data Input Screen When the VCCTL builds a digital microstructure based on input data measured from a cement, it also calculates a spatial distribution of the cement particles as isotropic two point correlation functions for each of the different phases. A two point correlation function i s statistical representation of the correlation between random variables at different points in space. In this case, the correlation function for each phase describes the probability that a pixel a

PAGE 46

46 specific distance away from the current pixel is of the sa me phase as the current pixel. When the s egmented image is measured to obtain phase fraction data, this data is also measured from the image. The VCCTL uses these two point correla tion functions to build a three dimensional microstructure that is consisten t with these functions. Figure 4 2 details the screen in which the phase and volume fractions can be input, as well as the two point correlation functions for each phase. Cement Particle Size Distribution Laser particle size analysis is used to obtain the particle size distribution curve that the VCCTL requires for cement. This information was obtained using a Horiba Laser Particle Size Analyzer, which measures the diffraction of light scattered by particle s as they pass through a laser beam. The machine uses compressed air to blow particles through the beam and sequentially measures the size of each particle. Particles are measured until the size distribution no longer changes. An example particle size dist ribution is shown in Figure 4 3 Figure 4 3 Cement Particle Size Distribution 0 1 2 3 4 5 6 7 8 9 0.1 1 10 100 1000 % Passing Particle Diameter (m

PAGE 47

47 Cement fineness is typically measured using the Blaine air permeability test, which is detailed in ASTM C 204. The test measures the fineness of Portland cement measuring the r ate at which air under suction (via a manometer) passes through a cement powder bed. The rate at which the air passes through the cement is compared to the rate of a known standard, and the specific surface, in m 2 /kg, is calculated via an empirical equatio n. While this method is useful for comparing the relative fineness of cements, it does not provide a par ticle size distribution, which t he VCCTL requires so that the digital microstructure will reflect the fineness of the input cement, which has a substant ial impact on the rate of hydration, with finer cements resulting in higher rates of hydration (Mindness & Young, 1981) This input is also selected on the screen shown in Figure 4 2. Cement Particle Shape Data Figure 4 4 Mi crostructure Simulation Parameters The VCCTL builds the digital microstructure using spherical particles, though it also has provisions for the use of real cement particle shapes. These particle shapes are obtained using X ray computed micro tomography sca ns of real cement particles, which are then individually

PAGE 48

48 modeled using a spherical harmonic expansion of the particle surface. The coefficients of the equations used to model the particle can then be used to reconstruct the particle virtually with arbitrar y size and orientation (Bullard & Garboczi, 2006) No cements were characterized in this way during this research, however the VCCTL cont ains a library of cement particle shape sets. A cement from the library with particle sha pes judged to be similar to that of the cement used in this research was used when simulations with real particle shapes were run. A screenshot showing this option is visible in Figure 4 4. Cement Particle Dispersion The VCCTL provides the option to disper se the cement particles when it builds the virtual microstructure. The functional effect of this is that space between particles is more uniform, which can result in better hydration and a corresponding increase in strength (Neville, 2011) This option is provided to simulate the mutual repulsion of cement particles that occurs (Mamlouk & Zaniwski, 2010) when water reducing admixtures are present. When simulating mixes with water reducing adm ixtures, this option (shown in Figure 4 4) was enabled and set to a value of one Binder System Size By default, the volume of the microstructure created by the VCCTL is 100 microns cubed. The size of the microstructure can increased or decreased by the us er, though a linear increase in a side of the microstructure will result in a cubic increase in volume, impacting the time required for the computer modeling of microstructure operations. Hydration Modeling Inputs The following inputs are used by the VCCTL to govern the behavior of the hydrating microstructure. Heat of Hydration as obtained via ASTM C 1702

PAGE 49

49 Curing Conditions Isothermal Conduction Calorimetry ASTM C 1702 ASTM C 1702 details a standard test method for the measurement of heat of hydration of hydraulic cementitious materials using isothermal conduction calorimetry, a process which has seen widespread adoption for cementitious research in the last few decades (Ferraro, 2009) The method details the construction and function of an isothermal calorimeter, the recommended procedures for the preparation of samples, and explains the analysis of the results obtained from a calorimeter to d etermine heat of hydration. The test method defines an isothermal conduction calorimeter as a constant temperature heat sink to which two heat flux sensors and sample holders are attached in a manner that insures good thermal conductivity. The two sample h olders are required for the measurement of one sample as the other holder carries a blank sample that evolves no heat. Figure 4 5 Simple Isothermal Calorimeter The heat of hydration evolved by the sample flows across the heat sensor and into the heat sink. The measured heat of hydration is the difference between the heat flow of the sample cell and the reference cell. A schematic diagram of a basic isothermal calorimeter is shown in Figure 4 5 (ASTM, C1702, 2009)

PAGE 50

50 Figure 4 6 Admix Ampoule (photo by author) One of the defining characteristics of hydraulic cements is that they exhibit an exothermic reaction when mixed with water. Ideally, the heat released from this reaction should be measured from the moment the sample is mixed with water. The sample should also be in thermal equilibrium with the environment within the calorimeter, as any heat flow through the sensor that exists prior to hydration can skew the measurements taken by the calorimeter. The test method details a procedure by which these two conditions can be met. An apparatus that holds the measured quantity of water in reserve is used to allow the sample to reach thermal

PAGE 51

51 equilibrium before mixing occurs. Th is apparatus also allows for the stirring of the mixture inside the calorimeter once the water is added. Such an apparatus, known as an Admix Ampoule, is shown in Figure 4 6. Per the test specifications, the measurement is started when the calorimeter shows no heat transfer between the sample and the calorimete r. The water is then added and mixed, and the heat evolution is measured for 7 days. The specification also outlines an alternative method for sample mixing, in which the sample is mixed outside the calorimeter and immediately inserted and measured. While much simpler than the previous procedure to execute, the test m ethod acknowledges a few shortcomings of this method (ASTM, C1702, 2009) The mixing of the sample prior to insertion in the calorimeter necessarily means that the e arliest stages of hydration. The baseline of the data may also be disrupted due to heat flow in or out of the specimen as it reaches equilibrium. For this research the internal mixing procedure was chosen, based on the acknowledged shortcomings of the exte rnal mixing procedure in the test specification, as well as past research showing the viability of the internal mixing procedure (Ferraro, 2009) The specific sample preparation procedure used for this research was as follows. The amounts of cement and water to be used in the sample vial were calculated such that the total heat capacity of cement and water together were equal to that of the reference sample. The cement is then carefully massed into the sample vials. The water wa s also carefully massed into the admix ampoule syringe. When the proper amount of water has been measured into the syringe, air was drawn into the syringe to empty the needle and create space. This helped to prevent leakage and premature hydration while th e sample was reaching thermal equilibrium within the isothermal calorimeter. At least three samples for each mix configuration was used, with two mixes measured during one 7 day test period. Vials and admix ampoules were

PAGE 52

52 assembled with a small amount of va cuum grease to seal connections and inserted into the calorimeter. For this research a Tam Air isothermal calorimeter was used, shown in Figure 4 3. This calorimeter consists of 8 channels, each channel consisting of two cells in the configuration shown pr eviously in Figure 4 7. Samples were allowed to equilibrate for at least 12 hours overnight. Measurement began when the water was injected into the sample vial. Samples were stirred manually for 60 seconds each, after which they were measured for seven day s Figure 4 7 Tam Air Isothermal Calorimeter with sample and admix ampoule loaded (photo by author) The resulting power curve for the seven day hydration reaction can be integrated over time to obtain a heat of hydration curve, which in turn can be input directly into the VCCTL as a text file. The input screen for the selection of a heat of hydration curve can be seen in Figure 4 8. For each computational cycle of the hydration reaction the VCCTL calculates the amount of heat released. Heat of hydrat ion data is not required by the VCCTL for hydration simulations, as the

PAGE 53

53 Figure 4 8 Hydration Simulation Input Parameters model assigns the same user input length to each cycle, however if the experimentally measured heat of hydration data (via ASTM C 1 702) is input into the VCCTL, the model will assign a length to each computational cycle of the simulation such that the simulated heat evolved during that cycle matches up with the input heat of hydration curve. Curing Conditions The VCCTL provides input options to simulate different possible curing conditions for the hydrating microstructure. The thermal conditions regulate the rate at which the heat evolved

PAGE 54

54 during hydration leaves the microstructure, with the isothermal option setting that rate to be equ al to the rate at which heat is evolved (keeping the microstructure at ambient temperature), the semi adiabatic option setting the rate at a user specified value, and the adiabatic option not allowing any heat to leave the microstructure at all. The options for saturation conditions influence the water available for the hydration simulation. The saturated option allows the microstructure to never run out of water for hydration reactions, while the sealed option limits the available water to that which is in the initial microstructure. For the purposes of this research, the saturated option was always selected. Aggregate Input Data Most of the input data required by the VCCTL for aggregates come from tests that are performed on aggregates as a matter of course, the results of which are readily available. A brief summary of these tests follows. ASTM C 136 Standard test method for the sieve analysis of fine and coarse aggregate. This test is run on all aggregates used at the FDOT ASTM C 127 Standard te st method for Density, Relative Density (Specific Gravity), and Absorption of Coarse Aggregate. This test is also run for all aggregates used at the FDOT The primary use of these inputs by the VCCTL is simply to convert the mass fractions to volume fracti ons when batching the virtual mix. Aggregate bulk and shear moduli are not among the data readily available to characterize aggregate. While these properties can be measured from rock cores sourced from the mine where the aggregate originated, there is a d egree of inherent variability to natural aggregate that makes obtaining representative cores difficult. This process was not performed for this research due to time and scope constraints. Instead, some values were assumed for the coarse and fine aggregate used in this study based on the knowledge that the VCCTL requires this data for the calculation

PAGE 55

55 of the modulus of concrete. As concrete is mostly aggregate by volume, bulk and shear moduli will potentially have a substantial influence on the results. Thus, future research should include a detailed investigation of these properties. The VCCTL accounts for aggregate shape data as a statistical distribution of spherical harmonic coefficients, which is obtained via X ray computed tomography. The VCCTL includes shape data for a variety of aggregates in its built in library of materials, and the influence of aggregate shape on simulation results was investigated as part of this study and is discussed in Chapter 7.

PAGE 56

56 CHAPTER 5 SEM MICROANALYSIS Overview The phase volume fraction and surface area fraction are both required inputs for VCCTL to simulate cement microstructure as it hydrates and are acquired via a combination of scanning electron microscopy and image analysis. The process was orginally developed at NIS T and consists of three primary operations. 1. Backscatter Electron and X ray Map Image Aquisition 2. Image Processing 3. Creation and Analysis of Segmented Image Before any of these operations can take place a significant amount of sample preparation is required to enable the imaging of the cement being analyzed within the scanning electron microscope. Sample Preparation The phase volume and surface area fraction data required by the VCCTL is obtained through a complex process involving image acquisition using a scanning electron microscope (SEM). Before this process can even begin however, the cement begin analyzed must b e prepared so that imaging is possible. The analysis process looks at cross sections of cement grains which must be perfectly flat and free of scratches, tear out and other imperfections that could skew measurement or cause imaging artifacts. The process r equired to achieve this is complex and sensitive. The original procedure for the preparation of cement powder specimens for imaging in this way was developed by NIST to provide inputs for different cement hydration models. This original process, outlined in the next few paragraphs, was used as a starting point for the procedure developed for this project.

PAGE 57

57 The cement is first mixed with an optical grade epoxy resin in proportions that result in a stiff paste. The paste is then formed into a ball and pressed into a specimen mold. The mold consolidated by tapping them mold sharply upon the table top. Once consolidation is complete, specimen is then cured according to the guidelines recommended by the epoxy manufacturer. After the specimen has been cured, a fre sh surface is exposed by grinding or sectioning with a wafering saw. Following the resin impregnation process, the specimen is ground with silicon carbide sandpaper using progressively finer abrasives, stopping at 600 grit. This proc ess serves to remove th e damage to the surface left by the tearing action of the diamond saw blade, and prepares the surface for the final polishing steps. The specimen is polished using progressively finer grit diamond abrasive and a polishing cloth, specifically 6, 3, 1, and 25 m. Each stage is polished until the imperfections left from grinding with the preceding abrasive are removed. Ethanol based lubricants are used throughout the cutting, grinding, and polishing process in order to prevent further hydration of the cement grains or paste. The final polished surface is carefully cleaned, after which a carbon coating is applied to render the sample conductive for imaging in the SEM Improvements The original procedure developed by NIST was almost 20 years old when this resear ch project began. Though the fundamental process remains the same, substantial improvements to the speed of preparation and consistency of quality were made. One of the biggest improvements made to the existing procedure was the automation of the grinding and polishing operations. The original process involved hand grinding of samples using silicon carbide paper on plates of glass, followed by hand polishing us ing a manual lapping wheel. These operations are very labor intensive and require a degree of operator skill.

PAGE 58

58 Figure 5 1 Saphir 550 Semi Automated Grinder Polisher (photo by author) An automated grinder polisher was acquired for this project specifical ly for the purpose of improving sample quality and reducing preparation time. Shown in Figure 5 1, the Saphir 550 grinder polisher allows for precise control of all the variables involved in the process of grinding and polishing samples. The basic operatio n of a polisher is simple. A working wheel with some type of abrasive, either a silicon carbide paper, a resin bonded diamond disc, or a polishing cloth with a diamond suspension, spins at a user specified RPM. The specimen to be polished is then held agai nst the working wheel as it spins and moved in a circular motion. In the case of a manual grinder polisher, the user would be in charge of holding the specimen against the working wheel. This semi automated grinder polisher, however, has a powerhead in whi ch up to six two inch specimens can be mounted. This powerhead also spins at a user specified RPM. In the case of this polisher, the powerhead can spin from 30 to 150 rpm, and the working wheel can spin from 50 to 450 rpm.

PAGE 59

59 When grinding and polishing manu ally, the user is also responsible for the amount of force with which the specimen is pressed against the working wheel. The Saphir 550 allows each of up to six specimens to be pressed against the working wheel with a user specified force ranging from 5 10 0 N. In addition, the user can specify a two stage force profile, wherein the force applied is one value for initial period, ramps up to a higher value for the majority of the operation, and then falls back off to the first value for the final stages of th e operation. This allows for a period of lighter material removal at the end of a grinding or polishing stage, resulting in a better surface finish. The user of a grinder polisher is also responsible for maintaining the flatness of the surface being polish ed. The Saphir 550 maintains the flatness of the specimens through the use of an automated powerhead and individual specimen forces. A typical grinding and polishing operation consists of several stages, using progressively finer abrasives in each stage. D number of grinding operations may be used. When the surface has been ground to an acceptable finish, the polishing operations commence. Using a polishing cloth and diamond abrasive, the sp ecimen is polished using progressively finer abrasives until the surface finish reaches the desired level of smoothness. When using a manual grinder and polisher, the user is responsible for all aspects of the operation, including the application of lubric ant to the working wheel, determination of when to move to the next abrasive stage, how much abrasive to use, and how much force to apply during each stage. The advantage of a semi automated grinder polisher is that once the ideal series of grinding and po lishing operations have been determined, they can be programmed into the grinder polisher. The Saphir 550 will perform all of the tasks required

PAGE 60

60 during a grinding and polishing operation except for the changing of the working wheel when switching to a fine r abrasive. During the development of the grinding and polishing procedure, inspection of the surface finish of the specimen after each step was useful to determine its efficacy. An optical microscope, purchased as part of this project, was used for this purpose. The microscope used for this purpose was a Nikon AZ 100, which is shown in figure 5 2. With maximum resolution of 0.5 microns and a magnification of up to 650x, this microscope allowed for the evaluation of the flatness and surface quality of samp les after each polishing step. Figure 5 2 Optical Microscope (photo by author) Some experimentation was necessary to develop a grinding and polishing process that resulted in specimens of suitable quality. There were initial issues with cement particles being

PAGE 61

61 torn from the resin during the grinding steps. This was solved by changing the initial grinding media, as well as by switching to a harder resin. A resin bonded diamond abrasive d isk was used initially, but it was found that the hardness of the diam ond abrasive left deep scratches in the specimens that could not be removed during the later polis hing steps. The solution was to switch to silicon carbide grinding papers. These papers offer a more progressive grind, as the softer silicon carbide actually becomes finer in grit as the paper is used. This proved much more effective. Other variables of the grinding and polishing process were experimented with, including specimen force, lubricant and abrasive dispensing intervals, and head and wheel speeds Th e eventual procedure that was settled on is summarized in T able 5 1. Table 5 1 Grinding and Polishing Procedur e The final step to specimen preparation was to deposit a conductive coating on the surface of the sample. For this purpose, an evaporative carbon coater was used, shown in Figure 5 3. This Carbon coater used a high electrical current combined with a sharpened graphite rod to coat a specimen in a nanometer thick layer of el emental carbon. The specimen is placed in the coating chamber, where a vacuum is created and the current is then passed through the carbon rod, causing the carbon to evaporate and be deposited on the specimen. Specimens were coated Step Time Abrasive Abrasive Type Media Force Head Speed (rpm) Wheel Speed (rpm) Lubricant Interval Abrasive Interval 1 1:00 600 Silicon Carbide Grinding Paper 20 N 150 149 0:30 2 1:00 1200 Silicon Carbide Grinding Paper 20 N 150 149 0:30 3 1:00 4000 Silicon Carbide Grinding Paper 20 N 150 149 0:30 4 2:00 6 m Diamond Suspension Silk Cloth 15 N 150 149 0:45 0:45 5 2:00 3 m Diamond Suspension Silk Cloth 15 N 150 149 0:45 0:45 6 2:00 1 m Diamond Suspension Silk Cloth 15 N 150 149 0:45 0:45 7 4:00 0.25 m Fumed Silica Silk Cloth 10 N 150 149 0:30 2:00

PAGE 62

62 for 4 seconds, three tim es, which was found to produce a coating thickness that enabled effective imaging. Figure 5 3 Evaporative Carbon Coater (photo by author) Backscatter Electron and X ray Map Image Acquisition After the completion of the sample preparation process an unhydrated cement sample is analyzed using a scanning electron microscope. The two primary imaging modes used in this process are Backscatter Electron (BSE) and Energy dispersive X ray (EDS) mapping The BSE mode results in an image wherein the contrast is determined by the relative average atomic number (Z) of the different phases, with the brightest phases having the highest average. The image in Figure 5 4 demonstrates this. The brightest phase pr esent is the Ferrite phase (C 4 AF), due to the presence of Iron, with is denser relative to the other phases composed of mainly Calcium, Aluminum and Silicon. The darkest phase is the epoxy substrate, as it is a plastic and thus composed primarily of relati vely low density carbon polymer chains. The phases in order of darkest to lightest (lowest density to highest density) are the void (epoxy), gypsum, C 3 A, C 2 S, C 3 S and C 4 AF.

PAGE 63

63 Figure 5 4 Cement Grain Close up (photo by author) Figure 5 4 is a close up of a polished cement grain. It is possible to discern the four primary cement phases with a reasonable degree of accuracy based on the relative grey levels and the morphology of the phases (aluminate is typically dispersed within ferrite, for instance) howev er, computerized quantitative analysis of the area and perimeter of each phase is not possible without more information, as different phases (belite and aluminate) fall within similar range of grey values. Figure 5 4 is provided to demonstrate the BSE imag ing mode, but the images used for analysis are typically of a larger area. An example of an actual image used for this process is provided in Figure 5 5. The second imaging mode used for this technique, EDS, is used to create an image that shows the locati on and relative concentration of specific elements within the cement sample.

PAGE 64

64 Figure 5 5 BSE image as used for phase analysis (photo by author) This is due to the characteristic nature of the X ray s emitted from the sample as they interact with the electron beam. An electron from the beam interacts with the electron cloud of an atom, ejec ting an inner shell electron. A n outer shell electron moves to the inner shell, releasing an X ray photon in the process. This photon has an energy that is character istic of the orbital and element from which it originated. By scanning each point of the image for X ray An example of an EDS elemental map is shown in Figure 5 6. The areas that are not black denote the pr esence of sulfur. This map is of the same region as Figure 5 5.

PAGE 65

65 Figure 5 6 Elemental Map for Sulfur (photo by author) This map, along with maps for Aluminum, Sodium, and Potassium, provides the additional information to distinguish the phases present in the BSE image that would otherwise have overlapping grey values. Figure 5 7 Maps required to distinguish Alite, Belite, Aluminate, Ferrite, and Gypsum (photo by author)

PAGE 66

66 The images shown in Figure 5 7 are those required to distinguish the 4 primary p hases in cement as well as gypsum. They are, from left to right, the BSE image, the Sulfur map, and the Aluminum map. Image Processing Figure 5 8 Phase Classification (photo by author)

PAGE 67

67 To combine the images shown in Figure 5 7, the program Multispec is used. Multispec is a program that has been developed at Perdue University for the purpose of processing and analyzing multi and hyperspectral images. A multispectral image is created by combining different imaging modes of the same area. For instance, a multispectral image of the same object could be some combination of thermal imaging, visual imaging, and infrared imaging. By assigning the BSE image and each EDS map to a specific color channel (Red, Green, Blue) a multispectral image of the cement pow der is created. This multispectral image contains more information on phase composition than any of the individual images that it was created from. Figure 5 8 illustrates that by assigning BSE image to the Red Channel, the Aluminum map to the Blue channel, and the Sulfur map to the Green channel, it is easier to discern the different phases of Portland cement in comparison to the grayscale image created by BSE imaging (Figure 5 4). Multispec has the capability to automatically analyze and assign a specific phase value to each pixel in the image using user defined training fields. The fields are visible in Figure 5 8, and are regions selected by the analyst that contain a specific phase. These user selected regions provide a reference for the program which it can apply in the classification of the entire image. The selection of these training fields is critical to the proper classification of the phases, and requires substantial practice. Once training fields have been selected, the program classifies the ima ge, using the settings displayed in Figure 5 9. The result is then saved to a .GIS file. A portion of the results provided by the Multispec program includes a statistical analysis and the calculation of the accuracy of the classification. The statistical r esult is based on the reference fields and the reliability of the classification, as well as the distribution of each of the phases being analyzed. These results for this image set are shown in Figure 5 10.

PAGE 68

68 Figure 5 9 Image Classification Dialog Figur e 5 10 Statistical Analysis of Classification

PAGE 69

69 Creation of Segmented Image When the selection of fields and classification of phases has been performed such that it .GIS file is viewed. The result is a false color image in which each color represents a different phase. Figure 5 11 is the image resulting from the classification of the field set being used in this example. This is very close to the final segmented imag e, and represents the end of the segmentation process itself. Figure 5 11 Segmented Image In order to take measurements of the segmented image, some final post processing is

PAGE 70

70 operation. This operation looks at a pixel and compares it to all the pixels within a user defined surrounding radius. This algorithm allows for the removal of fine noise without distorting or blurring the boundaries between different reg ions. This is critical as the surface area fraction measurement can be skewed by excessive noise in the segmented image. Figure 5 12 Segmented Image Prior to Thresholded Blur Operation Figure 5 12 shows the image prior to the application of the threshol ded blur filter. The settings for the filter are varied with using the preview function to remove the noise without removing an excessive amount of detail from the image. The only value that must remain the same is the threshold value, which is held at 1.

PAGE 71

71 Figure 5 13 Segmented Image after Blur Figure 5 13 shows the result of the thresholded blur. There is a substaintial reduction in noise and a general improvement of the distinctions between the different phases. Figure 5 13 is a typical resultant image from which area and perimeter measurements of the Portland Cement in which each pixel is represented by a numeric value correspo nding to the chemical phase of that pixel. This image is then used as the input for an executable provided by NIST that measures the volume and surface area fraction of each phase, and also determines the relative spatial correlations of each phase. Volume and surface area fractions are mea sured as area and perimeter

PAGE 72

72 fractions, as for isotropic systems (in which no preferential orientation is observed) the two parameters are the same (Bentz, Haecker, Feng, & Stutzman, 2002) The same is true for the spatial corre lation functions. The program provides an output as set of files that can be archived as a .zip file and used for input directly into the VCCTL for analysis. One aspect of this process that is important to consider is that each image set created represen ts what is a very small cross section of a larger bulk material. In order to obtain statistically valid data for input into the VCCTL the process must be repeated for many image sets. The results of each image set are then averaged to obtain a more represe ntative dataset. The following data is the result of the analysis of 14 fields including the above example and is provided in Appendix B Table 5 2 Volume and Surface Area Fraction Averages Phase Volume % SA % C 3 S 50.72 32.43 C 2 S 27.94 44.04 C 3 A 8.620 13.73 C 4 AF 12.71 9.781 Table 5 3 Standard Deviations of Volume and Surface Area Fraction Measurements Phase % % C 3 S 6.62 6.92 C 2 S 4.57 6.48 C 3 A 3.81 7.20 C 4 AF 3.59 4.88 The results presented in Table 5 2 provide a summary of volume and surface area fractions for the cement used in this study. The data displayed in Table 5 3 provides a summary of the standard deviation of the surface area and volume fractions for each phas e. The range standard deviation of the data obtained for the surface area fraction for each phase is between 3.0 and 7.5%. Some deal of variability between the fields is to be expected since Portland cement is a derivative of a naturally mined material tha t has intrinsic variability. However, since the

PAGE 73

73 process specimen and image preparation is rigorous, it is possible that the process can be refined further with better sample preparation. It is expected that acquiring a better final polish, as well as by c omparing the data to Powered XRD by back calculating volume percent from mass percent using the specific gravity values for each phase from literature will provide less variability in the phase acquisition. The refinement of this process is an ongoing effo rt, and as it is refined the simulations should more closely represent the actual reality of the concretes being simulated. Automated Cement Characterization Figure 5 14 Fields Analyzed by UF and RJ Lee Group (photo courtesy of April Snyder) The process of acquiring fields and processing them to create a segmented image is very labor intensive and has a fairly steep learning curve. In addition, multiple fields must be collected to have statistically valid data for the sample. Work was performed in cooper ation with RJ Lee Group to automate this process, with purpose of reducing the time and operator skill required to obtain cement images.

PAGE 74

74 A preliminary comparison study was conducted to investigate the potential for automated SEM field acquisition to reduc e time and labor requirement for the process. The first part of this study was the acquisition of 15 fields (BSE and EDS maps) of a polished cement powder mount manually using the SEM equipment at the University of Florida. Upon completion of the analysis at the University of Florida, the specimen was then sent to RJ Lee group, and the same fields were relocated and acquired using their automated SEM equipment. The fields acquired are shown in figure 5 14. The acquisition of the BSE image and EDS maps at th e University of Florida required approximately one hour of analyst time per field, with approximately 15 minutes of that hour being devoted to stage movement, refocusing, and BSE image acquisition, while the remainder of the hour was required to obtain the EDS maps themselves. This resulted in a total of roughly 15 hours of analyst time required for the acquisition of one set of fields for a single Portland cement specimen. The use of an automated SEM at RJ Lee Group allowed all of the fields to be relocate d at once, and placed in a queue to be analyzed. This meant that the analyst could step away after the fields to be analyzed were specified, and return with all of the images acquired. This required three hours of analyst time, which is a reduction of 75%. Table 5 4 Average Difference Between two Image Sets Phase Area Fraction Perimeter Fraction C 3 S 0.38% 0.81% C 2 S 2.75% 6.54% C 3 A 2.60% 6.17% C 4 AF 0.53% 1.20% In order to ensure that the images were of equivalent quality and to prove that the acquisition of fields manually had no impact on the resulting image segmentation process, both

PAGE 75

75 sets of images were processed in the same way at the University of Florida. T he resulting difference between the two sets is described in Table 5 5 The data shown in Table 5 4 shows the average difference between the data presented in tables 5 2 and 5 3 and the data obtained from fields acquired at UF. The two data sets ended up agreeing fairly well, especially when the much higher resolution of the X ray maps provided by RJ Lee Group is taken into account. This combined with the inherently subjective nature of the segmentation process itself, means that the substantial reduction in acquisition time that results from the use of an automated SEM also results in effectively identical data. Figure 5 15 New false color image (photo courtesy of April Snyder) After the comparison study, efforts continued at RJ Lee Group to further automate the characterization process. The ultimate goal of the collaboration was to completely automate the

PAGE 76

76 scanning and processing of cement samples. As of this research, some furth er improvements have been made to the acquisition process, but the image processing is still done manually. The chief improvement to the acquisition process was the creation of an image that is already colored in such a way to allow all the major phases to be distinguished. An example image can be seen in Figure 5 15 Figure 5 16 After processing, ready for classification (photo by author) This image was supplied directly from RJ Lee, and eliminates the process of combining the X ray maps and backscatter electron images. Different colors are assigned to different elements, as shown in the key at the lower left of Figure 5 15, and the resulting shades indicate the relative mass percentages of the elements present. The best example of t his can be seen in the difference between alite and belite. Silicon is assigned to green, and Calcium is assigned to blue,

PAGE 77

77 and because belite (C 2 S) has a higher ratio of silicon to calcium, it shows up as a greener shade of blue green, while alite appears bluer. This image is cropped, levels are adjusted, and the thresholded blur filter described earlier is applied to reduce noise. The resulting image, shown in Figure 5 16, can then be classified in multispec immediately. Image analysis proceeds as usual af ter this image is loaded into multispec. Several images were processed in this manner, and while when comparing manual vs. automated acquisition of fields the phase fractions were averaged, for this round the data from each field was looked at individually and the field from which the measurement most closely matched the XRD data was chosen to use with for input to the VCCTL. The phase volume fraction data measured from the images was compared to the expected volume fractions based XRD data obtained from t he cement plant. Table 5 5 shows a comparison between the plant data and the data measured from the image shown in Figure 5 17. Table 5 5 XRD Cement Phase Fraction vs. SEM Microanalysis Measurements Phase Mass % (XRD) Volume % (XRD) Volume % (Measured) Su rface Area % (Measured) Alite 68.31 69.14 66.14 58.44 Belite 15.04 15.73 15.72 20.37 Aluminate 5.39 5.59 5.74 17.12 Ferrite 11.25 9.54 12.04 4.07 As Table 5 5 shows, the phase volume fraction data measured from Figure 5 17 is within 3% of the values obtained via XRD. These phase fraction data were used for all simulations of this cement.

PAGE 78

78 Figure 5 17 Final Cement Image (photo by author) The decision to pick data from an individual field over the average of several was influenced primarily by the availability of the mass fraction data from XRD. The XRD data showed that the earlier image analyses systematically overestimated the volume frac tions of certain phases, particularly alite and belite. Because accuracy of characterization is the ultimate goal for this technique, it was deemed acceptable to pick an individual field that matched the XRD data closely because of standardized and well es tablished nature of the XRD test.

PAGE 79

79 CHAPTER 6 VCCTL OUTPUT DATA Overview The outputs provided by the VCCTL can be divided in to two categories. There are properties that are computed and measured continuously as part of the hydration simulation, referred to as continuous measurements by the VCCTL, and there are outputs that are the results of additional simulations calculated using a snapshot of the hydrated microstructure as an input, which are referred to as periodic measurements by the model because the y can only performed on static microstructures that have been saved at user specified intervals. The following list shows all the potential outputs from the VCCTL Cement and hydration product phase volume fractions Solution Concentration of Sodium, Potassi um, Calcium, Sulfate Solution Activity of Sodium, Potassium, Calcium, Sulfate Total porosity Fraction porosity connected in the x, y, and z direction Average fraction porosity connected Fraction solids connected in the x, y, and z direction Average fractio n solids connected Solution pH Solution Conductivity Non evaporable water content (ignited and unignited) Gel/space ratio Binder temperature Heat release Degree of hydration Cycles run Elastic Modulus of the paste Elastic Modulus of the concrete Strength of the concrete Of these, only average fraction solids connected, modulus of concrete and strength of concrete were used in this research for reasons that will be discussed in the following pages.

PAGE 80

80 Continuous Measurements Figure 6 1 Continuous Measurements Display Page The VCCTL provides 64 continuous output parameters that describe the hydrating cementitious microstructure on the screen shown in F igure 6 1. Of these, 34 describe the volume fractions of the cement and hydration produ cts that make up the physical composition of the microstructure. The other outputs include solution concentrations and activities for sodium, potassium, calcium and sulfate, the total porosity, the fraction porosity and fraction solids connected in the x, y, and z directions, the average fraction porosity and solids connected, the pH and conductivity of the pore solution, non evaporable water content, gel/space ratio, temperature

PAGE 81

81 of the binder, heat release, degrees of hydration for cement and fly ash, and the number of simulation cycles run. These outputs provide a wealth of fundamental information about the hydrating cement paste. Some of these outputs, specifically those that describe the concentration, pH, conductivity, and ionic composition of the pore solution, are intended to provide information that can be linked to the durability of concrete via degradation modeling (Bullard J. W.) Some work has been done with ion chromatography of extracted pore solution to evaluate the a ccuracy of the (X. Feng; NIST; Northwestern University, 2004) with mixed results. The VCCTL itself however, provides no direct provision for the prediction of concrete durability or s ervice life. Many of the outputs provided have also been used for the experimental verification of the accuracy of the hydration model. The volume fraction predictions, for instance, have been compared to experimental values obtained via in situ X ray dif fraction techniques (Valentini, et al., 2013) The prediction of degree of hydration has been experimentally verified multiple times (Bentz D. P., 2006) (Bentz, Feng, Haec ker, & Stutzman, 2000) via measurement of non evaporable water. The heat release measurement provided by the model has also been used in this way (Bentz, Feng, Haecker, & Stutzman, 2000) through experimental measurements o f heat of hydration via ASTM C 186. A few of the measurements provided as continuous outputs have seen direct applications for prediction of concrete performance. The most important of these is the fraction solids connected value. Because this value descri bes the percentage of the solid microstructure that has percolated through the simulation volume over time, it can be used to describe the phenomena of

PAGE 82

82 matri x. Figure 6 2 Fraction Solids Connected Curve Experimental tests for the setting time of cement typically adopt arbitrary limits to define resistance of mortar t o penetration with a needle with a specific cross sectional area, and defines initial set as a resistance of 500 pounds per square inch, and final set as 4000 pounds per square inch. The values for defining initial set and final set for the fraction solids connected value can be assigned in a similarly arbitrary fashion. This method has been used in industry applications in combination to evaluate the effects of temperature on set time (Bullard, Garboczi, & Stutzman, 2013) with some success. For this research, the value for initial set was defined at a value of 0.2, and final set was defined at 0.8. An example of the fraction solids connected output from the VCCTL can be seen in Figure 6 2. 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0.00 2.00 4.00 6.00 8.00 10.00 12.00 Fraction Connected Elapsed Time (hours)

PAGE 83

83 Some work has also been performed with the prediction of the compressive strength of gel/space ratio theory (Garboczi, Bullard, & Bentz, 2004) This method is somewhat li mited, as it requires experimental measurements at early age to calibrate the empirical theory and enable later age predictions. It is also noteworthy that this is not a built in function of the VCCTL, but an application of one of is outputs. Periodic Meas urements Figure 6 3 Periodic Measurement Display Page

PAGE 84

84 The second category of measurements are those obtained from calculations run on hydrated microstructures of various ages. It is from these outputs that elastic modulus and strength properties are obtained. The measurements screen for periodic outputs is shown in Figure 6 3. Elastic Modulus The VCCTL computes the elastic moduli of both the hydrated cement paste and the mortar or concrete. The modulus of the paste is first computed from the hydrated microstructure using a finite element model. The resulting modulus of the paste is then used as an input for another, slightly different finite element calculation, which calculates the modulus of the concrete based on the computed paste modulus as well as the input bulk and shear moduli for the aggregates used. A summary of the methods by which the se computations are executed are summarized as follows. Hydrated Cement Microstructure Modulus The elastic modulus of hydrated cement paste is computed using an effective medium theory (EMT) for three dimensional spherical particles (Garboczi & Day, An algorithm for computing the effective linear elastic properties of heterogeneous materials: Three dimensional results for composites with equal phase ratios, 1995) This theory calculates the bulk and shear moduli of the composit e paste using known values of the bulk and shear moduli and volume fractions of each specific phase present in the paste. The equations used to calculate the effective shear and bulk moduli of the paste are as follows:

PAGE 85

85 Where G and G n K and K n are the shear and bulk moduli of the composite and n th phase respectively, nd c n is the volume fraction of the n th phase. An algorithm using these equations measures the composite elastic modulus of the three dimensional microstructure by treating the digital image as a mesh of linear finite element, where each voxel is an eleme nt. With individual phase properties supplied to each voxel, composite properties are computed by applying a strain and computing the appropriate energy averages (Garboczi & Day, An algorithm for computing the effective linear elast ic properties of heterogeneous materials: Three dimensional results for composites with equal phase ratios, 1995) Concrete and Mortar Modulus The resulting modulus of the paste is then used to calculate the modulus of the concrete. This calculation uses a more complex differential EMT, which models the concrete as a matrix with spherical inclusions. These inclusions are the aggregate particles, which are surrounded by a thin shell of altered matrix material (representing the interfacial transition zo ne), making them composites themselves. The fundamental principle of the technique used here is that the aggregates and surrounding shells are mapped to a single particle with a diameter equal to that of the outer shell, and with effective moduli calculate d based on the moduli of the particle and shell. The equations to perform this mapping are quite lengthy, and detailed in A ppendix A

PAGE 86

86 The new effective particles are larger than the original aggregate particles, and thus the original particle volume fract ions must be normalized to account for the change in diameter. The equations used to perform this normalization are as follows: Where and are the original and normalized volume fractions, and are the original and normalized partial volume fractions of the j th type of particles lying in a certain diameter range, is the diameter of the original particle, and is the diamet er of the j th type new effective particle. The resulting bulk and shear moduli can now be solved for numerically using a set of differential equations that are also detailed in A ppendix A This numerical method was verified using the same method by which t he cement paste modulus is calculated, that is a three dimensional digital image was built with the aggregates and their surrounding zones of altered matrix material, and the effective modulus of the image was found using finite element simulation. Excelle nt agreement was found between the two methods. (Garboczi & Berryman, 2001) Compressive Strength The VCCTL finds the compressive strength of concrete via a direct empirical relationship. The available literature on this relati onship points to the ACI 318 code as the origin of this equation (Garboczi, Bullard, & Bentz, 2004) From ACI 318 an equation for the

PAGE 87

87 calculation of compress ive strength can be found. When used with metric units, this equation take s the following form Solving this equation for compressive strength yields Or if the equation is rearranged The VCCTL source code however, shows a different relationship between strength and modulus, which takes the following form: Figures 6 4 through 6 6 show example strength and modulus data from simulations. In Figure 6 6 it can be seen that the relationship between strength and modulus is indeed that of the equation from th e source code The commentary relating to this equation within the source code also indicates that the strength values provided from this empirical relationship are for large cubes of unspecified size of the type commonly used in Europe. Literature on the relationship be tween concrete cylinder and cube strength (Elwell & Fu, 1995) indicates that the strength of cylinders is approximately 0.8 that of cubes, however for this study the output of the model will be compared directly to measured val ues as discussed in Chapter 8.

PAGE 88

88 Figure 6 4 Figure 6 5 Predicted Strength vs. time, w/c 0.4 0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 Young's Moduls (GPa) Days Simulated 0 5 10 15 20 25 0 5 10 15 20 25 30 Strength (MPa) Days Simulated

PAGE 89

89 Figure 6 6 Strength vs. Elastic Modulus, w/c 0.4 y = 0.0006x 2.9831 0 5 10 15 20 25 30 30.5 31 31.5 32 32.5 33 33.5 34 Srength (MPa) Elastic Modulus (GPa) Simulated

PAGE 90

90 CHAPTER 7 PHYSICAL TEST PROGRAM Physical Testing Program All experiments relied on physical data from real concrete to evaluate the effectiveness of the model. A summary of the different mixes used in this research can be seen in Table 6 1 Table 7 1 Mixture Design Summary The cement and aggregates used in all mixes were obtained from a single source. For the primary validation study to gauge the influence of the different inputs on simulation accuracy, a range of mixes with different water contents were made. These mixes al l had the properties listed above, with water to cement ratios of 0.4, 0.45, 0.5 and 0.55. The difference in the volume of water for each mix was accounted for by changing the total aggregate content, while keeping the ratio of coarse to fine aggregate the same. Physical specimens were also created with different dosages of two commonly used water reducing admixtures, WRDA 60, which is a Type D mid range water reducer and retarder, and ADVA 600, a Type F high range water reducer. The above mix at a w/c of 0 .4 was used for all test mixes with admixtures, with the 0.4 mix from the input response study acting as a control. Specimens were prepared following the standard procedures outlined in ASTM C 192. Prior to batching, an approximate amount of coarse aggrega te was bagged and soaked for at least 24 hours. The bags were pulled from the soak an hour prior to batching to allow excess water to Material Mix 1 (0. 4 w/c) Mix 2 (0.45 w/c) Mix 3 (0.5 w/c) Mix 4 (0.55 w/c) Mix 5 (0.4 w/c) Mix 6 (0.4 w/c) Mix 7 (0.4 w/c) Cement (lb/yd 3 ) 681 681 681 681 681 681 681 Water (lb/yd 3 ) 272 306 341 375 272 272 272 Fine Agg. (lb/yd 3 ) 1093 1061 1030 995 1093 1093 1093 Coarse Agg. (lb/yd 3 ) 1681 1631 1586 1532 1681 1681 1681 Admixture WRDA 60 Type D ADVA 600 ADVA 600 Dosage (oz/cwt) 27.2 13.62 27.2

PAGE 91

91 drain. While the coarse aggregate was draining, the cement and fine aggregate were weighed and put into buckets. The coars e aggregate was then batched, with a small amount of aggregate set aside from each bag. The aggregate set aside was weighed wet, then put into the oven overnight to remove all moisture following the procedures outlined in ASTM C 70. The following day the w ater content and coarse aggregate contents were adjusted based on the difference between the assumed moisture content and the results of the test. Prior to mixing a small batch approximately 10% the size of the primary batch was mixed. This served to coat the inside of the mixer with mortar, and prevent any mortar loss from the primary mix. Coarse and fine aggregates along with approximately half the water were then added to the mixer and mixed until evenly distributed. The cement and remaining water were a dded and mixed for three minutes. The concrete was then allowed to rest for approximately 3 minutes. Following this step the admixture (if any) was added and the concrete was mixed for another three minutes. All batches were tested for Slump (ASTM C 143), Unit weight (ASTM C 138), Temperature (ASTM C 1064), and Air content (ASTM C 173) in accordance with the ASTM standard method for each test. Mixes with admixture content also had a portion of their mortar sieved out and placed into molds for the time of s et test. The results for the air content, unit weight, temperature, and slump tests are shown in Table 6 2 Table 7 2 Summary of Test Results for Fresh Concrete Material Mix 1 Mix 2 Mix 3 Mix 4 Mix 5 Mix 6 Mix 7 Slump (in) 1.25 6.25 8.0 8.75 11.0 Temperature (F) 75 72 74 75 76 74 74 Air Content (%) 2.5 2.25 0.3 0.0 3.0 3.0 3.75 Unit Weight (lb/yd 3 ) 143.4 142.2 139.1 137.7 141.3 140.5 142.16

PAGE 92

92 Figure 7 1 Preparation of Cylinders (photo by author) Figure 7 2 Compressive Strength and Elastic Modulus Testing (photo by author) for the testing of compressive strength and modulus. Cylinders were filled in two lifts, with a

PAGE 93

93 vibratory table used for 15 seconds between each lift to consolidate the concrete. Cylinder preparation is shown in Figure 6 1. Cylinders were then capped and allowed to cure for 24 hours before the stripping of molds and the first strength and modulus tes ts. The time of set test was run in accordance with ASTM C 403. The mortar was allowed to sit for 2 hours, after which penetration resistance measurements were taken regular intervals. The apparatus for the penetration resistance test is shown in Figure 6 3. Hardened cylinders were ground flat on both ends and tested for compressive strength and elastic modulus in accordance with ASTM C 39 and ASTM C 469 respectively. Figure 6 2 provides an illustration of these tests. Strength and modulus values were obtai ned at 1, 3, 7, 14, and 28 days in order to provide a wide range of time scales for the evaluation of the accuracy of the VCCTL. Figure 7 3 Time of Set Test Apparatus (photo by author)

PAGE 94

94 Isothermal Calorimetry Isothermal calorimetry data was collected for each of the mixes that contained admixtures, as well as the control mix with the w/c of 0.4. For each cement, water, and admixture compound three samples were created and measured. Masses of cement and water for each sample were calculated such that the t otal heat capacity of the mixture was equal to that of the dummy sample. Admixtures and water were mixed by mass to reflect the actual proportions present in the concrete samples. The mixture was then loaded in to the Admix Ampule syringes by mass. Samples were allowed to equilibrate overnight prior to mixing, after which data was collected for 168 hours (7 days). A summary of tests and ages at which they were performed for all mixes is shown in Table 7 3. Table 7 3 Tests and Ages Test Mix 1 (0. 4 w/c) Mix 2 (0.45 w/c) Mix 3 (0.5 w/c) Mix 4 (0.55 w/c) Mix 5 (0.4 w/c) Mix 6 (0.4 w/c) Mix 7 (0.4 w/c) ASTM C 39 Compressive Strength Days 1, 3, 7, 14, 28 Days 1, 3, 7, 14, 28 Days 1, 3, 7, 14, 28 Days 1, 3, 7, 14, 28 Days 1, 3, 7, 14, 28 Days 1, 3, 7, 14, 28 Days 1, 3, 7, 14, 28 ASTM C 469 Elastic Modulus Days 1, 3, 7, 14, 28 Days 1, 3, 7, 14, 28 Days 1, 3, 7, 14, 28 Days 1, 3, 7, 14, 28 Days 1, 3, 7, 14, 28 Days 1, 3, 7, 14, 28 Days 1, 3, 7, 14, 28 ASTM C 403 Time of Set Day 1 Day 1 Day 1 Day 1 ASTM C 1702 Heat of Hydration Days 1 7 Days 1 7 Days 1 7 Days 1 7 Days 1 7 Days 1 7

PAGE 95

95 CHAPTER 8 COMPARISON OF PHYSICAL TEST DATA TO MODEL OUTPUTS Experimental Design The primary objective of the experiments conducted for this research was to validate the accuracy of the VCCTL in the simulation of the structural properties of concrete (as opposed to mortar or grout) and assess its potential value as a tool for the development of new concrete mixture designs. The vast majority of concrete mixes made today c ontain both supplementary some type of chemical admixture. While mixes containing these materials can be simulated by the VCCTL, the characterization of these materials requires the development and refinement of procedures at least as complex as those for Portland cement. The scope of this project precluded admixtures m ay be simulated was developed and investigated. Currently, the VCCTL does not possess the capability to properly model concrete mixtures which incorporate supplementary cementitious materials therefore, the concrete mixture designs considered for the valid ation study contained Portland cement as the only cementitious component. With this restriction imposed, the decision to test the accuracy of the simulations at a range of water to cement ratios was made. Water to cement ratio is widely known (Mindness & Young, 1981) to be the most important variable that affects the strength and elastic modulus of concrete. Concrete mixture incorporating water to cement (w/c) ratios of 0.4, 0.45, 0.5, and 0.55 were selected, with the lower boun d of 0.4 being the limit of feasible mixing without any water reducing admixture, and the upper limit of 0.55 being above the designed water content of most mixes used in industry applications (Florida Department of Transportation, 2014) The results of

PAGE 96

96 the experimental data of concrete mixtures created in the laboratory were compared directly to the compressive strength and compressive modulus elasticity data predicted by the VCCTL. It was determined that investigating the infl uence of each of the different material inputs accepted by the VCCTL would be important for evaluating its usefulness as a concrete mixture development tool. A control mix with a water to cement ratio of 0.4 was modeled first using all available input data then several simulations were performed on the same mixture, with a different non essential input removed in each simulation. Because compressive strength is calculated by the VCCTL from modulus based on a fixed empirical relationship, it was not conside red when comparing the influence of different input data. For the removal of inputs which influence the hydration simulation, changes to the fraction solids connected curve were also investigated. Experiments were conducted to investigate the accuracy of s imulation of some concrete mixtures containing water reducing admixtures. Water reducing admixtures typically increase the set time of fresh concrete due to the nature of the mechanism by which they operate, discussed in chapter 3. This effect can be measu red in fresh concrete using a simple time of set test, and can also be quantified using isothermal calorimetry. Because the VCCTL utilizes heat of hydration of Portland cement to calibrate the time scale of a simulation, it was decided to investigate the d ifference in set time would present itself in the model outputs when isothermal calorimetry data from a cement hydrated which incorporated a water reducing admixture. Additionally, it was postulated that the long term effect of the addition of a water redu cing admixture on strength and elastic modulus is negligible when mixes of the same water to cement ratio are compared (Mamlouk & Zaniwski, 2010) Given that this postulate holds true, this indirect method of accounting for adm ixture behavior could prove a substantial enhancement to

PAGE 97

97 the practicality of the VCCTL since, fraction solids connected value has been the the primary output of the model used to predict the changes in set time. Simulation Procedures Except for the mixes evaluating the sensitivity of the VCCTL to the absence of non essential inputs (defined in chapter 4), all mixes were simulated with all available input data. These inputs included cement phase volume and surface area fraction data, obtained via the SEM mi croanalysis method described earlier in this document, the mass fraction data for the different phases of gypsum, as obtained via XRD, particle size distribution curves, isothermal calorimetry curves, aggregate bulk and shear moduli, and aggregate shape da ta selected from the library of built in materials based on similarity to the aggregate used. Simulations were also performed with real particle shape data for the cement, which was selected from the material library within the VCCTL database, based on its resemblance to segmented imagery. To simulate the effects of water to cement ratio, the only changes made to each mix were the calorimetry curves, and the relative proportions of the mixes. Calorimetry data for the mix with 0.45 water to cement ratio was unavailable; the calorimetry data used for the mix at 0.4 water to cement was used as a substitute. The aggregate mass fractions, cement content, and water to cement ratio were modeled exactly as mixed. To gauge the sensitivity of the simulations to the d ifferent non essential inputs, several simulations with run with different inputs absent or disabled. All simulations used to evaluate input sensitivity were based on a control mix with a 0.4 w/c ratio. These included a mix with no calorimetry data, a mix with no real particle shape data, a mix with no aggregate shape data as well as a mix with a microstructure 50% larger (on a side) than normal. The influence of indiv idual simulations.

PAGE 98

98 For the investigation of set time simulation via isothermal calorimetry curves, four mixes were simulated with calorimetry data from three different admixture dosage combinations. One mix investigated the effects of a Type D (mid rage) water reducer and retarder, Grace WRDA 60, at a dosage of 4 ounces per hundred pounds of cement. Two mixes investigated the effects of a Type A high range water reducer, Grace ADVACAST 600, at dosages of 2 and 4 ounces per hundred pounds of cement. The fou rth mix was a control, with no admixture present. Isothermal calorimetry curves from each of these combinations were used in the simulations, and the mixes with admixtures present had cement particle dispersion enabled. Results Water to cement Ratio Study Figure 8 1 Compressive Strength vs. Time for different w/c ratios The physical test results of the four different w/c ratios are shown in Figures 8 1 and 8 2. 0 10 20 30 40 50 60 0 5 10 15 20 25 30 Strength (MPa) Days 0.4 w/c 0.45 w/c 0.5 w/c 0.55 w/c

PAGE 99

99 Figure 8 2 The data in Figures 8 1 and 8 2 show the measured strength and elastic modulus of the mixes increased as the water to cement ratio decreased. Figures 8 3 and 8 4 show that while there is very little apparent difference between the power output at the three different water to cement ratios, the ultimate energy reached at 7 days (7 day heat of hydration) increases slightly with an increase in water to cement ratio. Figures 8 5 through 8 12 show a comparison of the strength and modulus results obtained from ph ysical testing to the simulated results calculated by the model. The error bars for the physical test data represent plus or minus one standard deviation of the sample set for each measurement. The power and heat of hydration curves obtained via isotherma l calorimetry for the different w/c ratios are shown in Figures 8 3 and 8 4. 0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 Young's Moduls (GPa) Days 0.4 w/c 0.45 w/c 0.5 w/c 0.55 w/c

PAGE 100

100 Figure 8 3 Power vs. Time for different w/c ratios Figure 8 4 Energy vs. Time for different w/c ratios 0.00 2.00 4.00 6.00 8.00 10.00 12.00 14.00 0.00 20.00 40.00 60.00 80.00 100.00 120.00 140.00 160.00 Power (mw/g) Time (hours) 0.4 w/c 0.5 w/c 0.55 w/c 0.00 50.00 100.00 150.00 200.00 250.00 300.00 350.00 400.00 0.00 20.00 40.00 60.00 80.00 100.00 120.00 140.00 160.00 Energy (J/g) Time (hours) 0.4 w/c 0.5 w/c 0.55 w/c

PAGE 101

101 Figure 8 5 Young Modulus vs. Time, w/c ratio of 0.4 Figure 8 6 Compressive Strength vs. Time. w/c ratio of 0.4 0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 Young's Moduls (GPa) Days Measured Simulated 0 10 20 30 40 50 60 0 5 10 15 20 25 30 Strength (MPa) Days Measured Simulated

PAGE 102

102 Figure 8 7 Young Modulus vs. Time, w/c ratio of 0.45 Figure 8 8 Compressive Strength vs. Time. w/c ratio of 0.45 0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 Young's Moduls (GPa) Days Measured Simulated 0 10 20 30 40 50 60 0 5 10 15 20 25 30 Strength (MPa) Days Measured Simulated

PAGE 103

103 Figure 8 9 Figure 8 10 Compressive Stre ngth vs. Time. w/c ratio of 0.5 0 5 10 15 20 25 30 35 0 5 10 15 20 25 30 Young's Moduls (GPa) Days Measured Simulated 0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 Strength (MPa) Days Measured Simulated

PAGE 104

104 Figure 8 11 Figure 8 12 Compressive Strength vs. Time, w/c ratio of 0.55 0 5 10 15 20 25 30 35 0 5 10 15 20 25 30 Young's Moduls (GPa) Days Measured Simulated 0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 Strength (MPa) Days Measured Simulated

PAGE 105

105 A comparison of Figures 8 5, 8 7, 8 9, and 8 11 show that the difference between the simulated and me water to cement ratio increases. The simulated values overestimate early age modulus, while predicting later age modulus closely at lower w/c ratios. Simulated compressive strength is dramatically lower tha n measured compressive strength at all w/c ratios and ages. Some possible reasons for this will be explored in the discussion of these results. Input Sensitivity Study Figures 8 13, 8 14, and 8 15 show the effects of the removal of cement calorimetry data cement particle shape data, and aggregate shape data respectively, relative to both the simulated control with all available inputs and the measured values for said control. Figure 8 13 0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 Young's Moduls (GPa) Days Measured Simulated Control No Calorimetry Data

PAGE 106

106 Figure 8 14 Figure 8 15 0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 Young's Moduls (GPa) Days Measured Simulated Control No Cement Particle Shape 0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 Young's Moduls (GPa) Days Measured Simulated Normal No Aggregate Shape

PAGE 107

107 The absence of aggregate shape data and cement particle shape data (Figures 8 14 and 8 15) appear to have very li calorimetry data (Figure 8 13) has a more dramatic effect however. The simulation without simula tion with the calorimetry data, which over predicts modulus at early ages. At later ages both simulations under predicted the modulus, with the no calorimetry simulation doing so to a slightly larger extent. The effect of a simulation run with a microstru cture 50% larger on a side is shown by Figure 8 16. Figure 8 16 Like the simulations without aggregate or cement particle shape, the larger microstructure has very little influence on the simulated results. Of note is the increase in the time required to run a simulation of a larger microstructure. While the digital microstructure was only 50% larger 0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 Young's Moduls (GPa) Days Measured Simulated Normal Simulated Larger Microstructure

PAGE 108

108 on a side, the volume increase of the microstructure was cubic, resulting in a volume 3.375 times la rger than normal. This simulation took approximately 3.375 times longer to simulate than a microstructure two thirds the size. The influence of aggregate moduli is shown in Figure 8 17, with the results of simulations with three different aggregate moduli compared to the measured values. Figure 8 17 The influence of the aggregate of elastic modulus is simple and clear. Higher moduli aggregates result in higher moduli concrete in simulations. Three of the optional inputs examined in the input sensitivity study influence the formation or hydration of the virtual microstructure. Because of this, the effect of these inputs on the simulation of time of set via fraction solids connected was investigated. The results of this are shown in Figure 8 18. 0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 Young's Moduls (GPa) Days Measured 10 GPa 15 GPa (Normal) 20 GPa

PAGE 109

109 Figure 8 18 Fraction Solids Connected vs. Elapsed time Based in the values of 0.2 and 0.8 for initial and final set respectively, Figure 8 18 shows that the mix with no calorimetry data and the mix with no ceme nt particle shape data reach initial 30 minutes earlier and final set 30 minutes later than the control mix and the mix with the larger microstructure. The percolation of solids for the mix with no particle shape data does start and finish earlier than the mix with no calorimetry data, however, which takes much longer to reach the same values than the other mixes. The larger microstructure mix appears to be effectively identical to the control. Admixture Set Time Study One of the chief postulates of the adm ixture set time study was that water reducing admixtures do not have a significant influence on long term strength and elastic modulus. The measured values for strength and elastic modulus for the mixes with different admixture 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0.00 2.00 4.00 6.00 8.00 10.00 12.00 14.00 16.00 18.00 Fraction Connected Elapsed Time (hours) Control Larger Microstructure No Cement Particle Shape No Calorimetry

PAGE 110

110 compositions are shown in Fi gures 8 19 and 8 20 below. Error bars for the no admixture mix represent plus or minus one standard deviation of the sample data for that day. Figure 8 19 Measured Strength vs. Time for Different Admixtures Figure 8 20 me for Different Admixtures 0 10 20 30 40 50 60 0 5 10 15 20 25 30 Strength (MPa) Days No Admixture 4 oz/cwt WRDA 60 2 oz/cwt ADVA 600 4 oz/cwt ADVA 600 0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 Young's Moduls (GPa) Days No Admixture 4 oz/cwt WRDA 60 2 oz/cwt ADVA 600 4 oz/cwt ADVA 600

PAGE 111

111 Figures 8 19 and 8 20 show that the mix with 4 ounces per hundredweight cement has a consistently higher strength and modulus at all ages. The other mixes agree more closely with the control mix, particularly as they reach 28 da ys. It is possible that the differences in strength and modulus are reflected in the simulations due to the dispersion of particles in the virtual microstructure for mixes with admixtures present. Figures 8 21 through 8 26 compare the results for strength comparison of the control mix, refer to Figures 8 5 and 8 6. Figure 8 21 Figures 8 21 and 8 22 show that the model predicts the modulus at later ages to within plus or minus one the standard deviations of measurements taken at those ages. The model does underestimate the modulus for the mix with 4 ounces per hundredweight cement of ADVA 600 (Figure 8 23). This corresponds with the increase in modulus when compared to the control mix in Figure 8 20. Once again, compressive strength is dramatically underestimated by the model in all cases. 0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 Young's Moduls (GPa) Days Measured Simulated

PAGE 112

112 Figure 8 22 Figure 8 23 me, 4 oz/cwt ADVA 600 0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 Young's Moduls (GPa) Days Measured Simulated 0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 Young's Moduls (GPa) Days Measured Simulated

PAGE 113

113 Figure 8 24 Strength vs. Time, 4 oz/cwt WRDA 60 Figure 8 25 Strength vs. Time, 2 oz/cwt ADVA 600 0 10 20 30 40 50 60 0 5 10 15 20 25 30 Strength (MPa) Days Measured Simulated 0 10 20 30 40 50 60 0 5 10 15 20 25 30 Strength (MPa) Days Measured Simulated

PAGE 114

114 Figure 8 26 Strength vs. Time, 4 oz/cwt ADVA 600 The power and heat calorimetry curves used as inputs to simulate the mixes with admixtures are shown in Figures 8 27 and 8 28. Figure 8 27 Power vs. Time 0 10 20 30 40 50 60 70 0 5 10 15 20 25 30 Strength (MPa) Days Measured Simulated 0.00 2.00 4.00 6.00 8.00 10.00 12.00 14.00 16.00 18.00 20.00 0.00 5.00 10.00 15.00 20.00 25.00 30.00 35.00 Power (mw/g) Time (hours) No Admixture 4 oz/cwt WRDA 60 2 oz/cwt ADVA 600 4 oz/cwt ADVA 600

PAGE 115

115 Figure 8 28 Energy vs. Time Both the heat and power curves show different degrees of delay for the second peak that marks the start of the hydration of C 3 S. The mix with n o admixture starts first about two hours after the introduction of the water, followed by the mixes with 4 oz/cwt WRDA 60 and 2 oz/cwt ADVA 600 at about three hours. The mix with the 4 oz/cwt dose of ADVA 600 starts last at about 4 hours after mixing. The delay remains fairly constant for approximately 20 hours. Figure 8 29 shows the fraction solids connected curves for each of the mixes as obtained fro m their respective simulations. The curves in Figure 8 29 match the isothermal calorimetry data both in o rder and in relative delay. The first mix in which the solids percolate is the control, with both the WRDA 60 and 2 oz/cwt ADVA 600 mixes following a little more than an hour later. The final mix to begin percolating is also the 4 oz/cwt ADVA 600 about an hour after the middle two. This also matches the heat and power curves from isothermal calorimetry. 0.00 50.00 100.00 150.00 200.00 250.00 300.00 0.00 10.00 20.00 30.00 40.00 50.00 60.00 70.00 Energy (J/g) Time (hours) No Admixture 4 oz/cwt WRDA 60 2 oz/cwt ADVA 600 4 oz/cwt ADVA 600

PAGE 116

116 Figure 8 29 Fraction Solids Connected vs. Time Figure 8 30 Penetration Resistance vs. Time, Different Admixtures and Dosages Figure 8 30 shows the results of time of set via penetration resistance testing for the different mixes. The time of set results for the two mixes with ADVA 600 correlate with the 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0.00 2.00 4.00 6.00 8.00 10.00 12.00 Fraction Connected Elapsed Time (hours) Control 4 oz/cwt WRDA 60 2 oz/cwt ADVA 600 4 oz/cwt ADVA 600 0 1000 2000 3000 4000 5000 6000 1:00 2:00 3:00 4:00 5:00 6:00 7:00 8:00 Penetration Resistance (psi) Elapsed Time (hours) Control 4 oz/cwt WRDA 60 2 oz/cwt ADVA 600 4 oz/cwt ADVA 600 Final Set Initial Set

PAGE 117

117 curves in Figures 8 29, 8 28 and 8 27, with a delay of about one hour for the mix with 2 oz/cwt, f ollowed by the 4 oz/cwt mix approximately one hour after. There is a discrepancy in the behavior of the mix containing WRDA 60 however. In the isothermal and fraction solids connected data this mix approximately follows the mix with the lower dosage of ADV A 600, setting/reacting between the control mix and the mix with the higher dose of ADVA 600. In this case, the WRDA 60 mix was the last to set, doing so approximately an hour after the higher dosage of ADVA 600. Discussion of Results Modulus and Strength The results of the water to cement ratio study mixes and simulations indicate that the VCCTL is a decent predictor of modulus at a water to cement ratio of 0.4, with progressively more overestimation of modulus as water to cement ratio increases. Figure 8 31 provides a visual representation of this overestimation. Figure 8 31 Simulated vs. Measured Elastic Modulus 0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40 Simulated Elastic Modulus (GPa) Mesured Elastic Modulus (GPa) 0.5 w/c 0.55 w/c 0.45 w/c 0.4 w/c

PAGE 118

118 The simulations of elastic modulus also overestimate early age values even when accurately predicting 28 day values. This effect is most evid ent in the mixes with a water to cement ratio of 0.4. The most obvious discrepancy between simulated and measured values is those of strength. Strength is dramatically underestimated by the VCCTL in all cases. Simulated modulus values are used to predict strength using a simple empirical relationship. Figure 8 32 shows the relationship between modulus and strength for both simulated and measured values. Figure 8 32 Strength vs. Elastic Modulus with VCCTL Power Fit and ACI 318 The predicted strength dat a from the simulations agrees almost with the equation found within ACI 318 almost exactly, as shown by the best fit power curve plotted in figure 8 32. In the commentary of the code it states that this equation originates from work done by S. Pichet of th e Siam Cement Group, another member of the VCCTL consortium, however no publications y = 0.0504x 1.9455 y = 0.0006x 2.9674 0 10 20 30 40 50 60 70 80 0 5 10 15 20 25 30 35 40 45 Srength (MPa) Elastic Modulus (GPa) Measured Simulated ACI 318 Power (Measured) Power (Simulated)

PAGE 119

11 9 exist detailing the materials used in concrete testing program from which this equation was developed. Other empirically based relationships for strength do exist however If the relationship derived from the ACI 318 code is plotted on Figure 8 32, it appears to agree quite well with the physically measured data. If the power best fit curve for the data is plotted, the resulting equation is When this equation is compared to the relationship derived from the empirical equation found in ACI 318, which takes the form: the two equations agree very closely. An example of the predicted strength from this equation compared to the embedded prediction in the VCCTL is shown in Figure 8 33. Figure 8 33 Strength vs. Time, Different Empirical Predictions 0 10 20 30 40 50 60 0 5 10 15 20 25 30 Strength (MPa) Days Measured VCCTL Embedded ACI 318 Predicted

PAGE 120

120 Though the relationship defined in ACI 318 appears as though it would predict strength fairly well at later ages in this circumstance, the use of empirical relationships do not take into account many of the factors that influence the behavior of concrete. It is possible that the these factors, the mos t likely of which being the scenario in which the coarse aggregates being modeled do not have the same properties as those of the materials used to develop the empirical relationship. Due to the lack of information on the materials used to develop the empi rical relationship between modulus and strength, any possible explanation for the drastic underprediction of strength for the concrete used in this study must come from specific knowledge about the properties of the materials used as well as an assumption about the nature of the materials used by Siam Cement Group. The concrete in this study exhibited fracture behavior that does not behave in the same fashion as the commonly described mechanism of failure for concrete. Typical concrete fails in compression when microcracks imitated at the interfacial transition zones surrounding aggregates propagate through the specimen, resulting in failure planes that pass around aggregate particles (Maso, 1996) In the concrete used in this st udy however, the failure planes passed through the aggregate particles. Figure 8 34 shows an example of a failed cylinder that exhibits this behavior. Coarse aggregate particles that have fractured are clearly visible in the two fragments shown in Figure 8 33. An aggregate particle face in one fragment is mirrored in the other fragment, which came from the opposite side of the failure plane. The conclusion that results

PAGE 121

121 from this is that the strength of the interfacial transition zone is greater than that of the aggregate particles themselves for the concrete used in this research. Figure 8 34 Crushed Cylinder Fragments (photo by author) This can be explained either by a better bond between the aggregate paste, weaker aggregate, or some combination of the se two factors. The coarse aggregate used in this study was observably porous and angular, with an absorption of approximately 4.5%. The diffusion of hydration products into porous aggregates which are initially saturated (as all aggregate was for this res earch) as well as the mechanical interlock between the cement paste and irregular aggregates surfaces can alter the interfacial transition zone to the point where it is no longer vulnerable to microcrack propagation (Maso, 1996) It has also been observed petrographically that the rough surfaces of Florida limestone aggregates provide good mechanical interlock and exhibit chemical bonding with cement paste (McClellan, Eades, & Fountain, 1993)

PAGE 122

122 The VC behavior of Florida limestone aggregate through an understanding of elastic modulus testing, and the assumption that the aggregate used by Siam Cement Group to develop the strength prediction formula exhibited typical interfacial properties. Because modulus of concrete is measured within the linear elastic range before microcracks begin to form and aggregate interfacial transition zone (ITZ) becomes relevant ( Neville, 2011) it is possible to have two concretes with the same modulus, even if the interfacial properties of the aggregates in those concretes are dramatically different. It follows that if given two concretes with the same elastic modulus, one w ith a typical aggregate interfacial zone and one with an interface similar to that which forms with Florida limestone, the concrete with typical aggregate interfacial behavior would develop microcracks more rapidly, resulting in earlier compressive failure This is the most practical explanation for within one standard deviation of sample data. Model Sensitivity The results of the investigation into the sensitivity of the model to the presence of different material input data show that several of the non essential input values have little influence on the outcome of simulations. Aggregate shape data, while documented to have an influence on the strength of concrete (Jones & Kaplan, 1957) has no effect on the simulated strength because it has no effect on elastic modulus, the value from which strength is calculated. The size of the simulated microstructure had a very slight effect on the resu lts of simulations in this case, though even a small increase in size results in a dramatic increase in simulation computing time. The influence of aggregate elastic properties on the elastic properties of the concrete is substantial, as the aggregate mak es up a large fraction of the concrete by volume, and the

PAGE 123

123 VCCTL calculates the modulus of the concrete based on the computed modulus of the paste as well as the relative volume fractions and elastic properties of the coarse and fine aggregate. The absence of isothermal calorimetry data resulted in a large underestimation of early age modulus and a small underestimation of late age modulus, as opposed to an overestimation of early age modulus and an accurate prediction of late age modulus when calorimetry da ta was present. The set time based on fraction solids connected also began earlier and lasted longer when no isothermal calorimetry data was present. The absence of cement particle shape data had no appreciable influence on modulus, but had a similar effec t on set time as the absence of calorimetry data. It is possible that if both of these inputs were missing the effect on set time could be compounded. Admixture Influence on Set Time The simulation of the delay in set time due to admixtures by using isoth ermal calorimetry data obtained from those admixtures was a partial success. The predicted delay in set time of the mixes containing two different dosage rates of ADVA 600 matched up with the measured set time delay of the actual concrete. The mix containi ng WRDA 60 did not however, with the predicted set occurring in roughly half the time of the actual measured set. Heat evolution and set time of concrete are two individual phenomena, and while they may be linked, the beginning of one does not necessarily imply the start of the other.

PAGE 124

124 CHAPTER 9 CONCLUSIONS The research described in this document led to several conclusions, which are described as follows. T he VCCTL is an effective tool for the simulation of elastic modulus of Portland cement concrete, pro vided the materials being simula ted are properly characterized. The VCCTL does not however, effectively predict the compressive strength of Portland cement concrete using Florida limestone with the inputs currently available. The use of heat of hydration data obtained from a mix containing an admixture as an input to the VCCTL does not guarantee accurate prediction of set time for a mix containing the same dosage of the same admixture. More work is needed to incorporate admixtures into the VCCTL reliably The process of acquiring cement phase volume and surface area fraction data has been improved substantially through the use of automated scanning electron microscopy and has resulted in a more efficient process for cement characterization for use in VCCTL. The empirical predictions in the VCCTL for compressive strength based on elastic modulus developed using ordinary coarse aggregates do not work for Florida limestone due to differences in concrete. More study is needed to accurately predict compressive strength based on the elastic properties and fracture behavior of concrete which incorporates these materials. A combination of isothermal calorimetry data and time of set testing has shown that ADVA 600 Type A superplasticizer delays set and heat release by the same amount of time. At the dosages explored within this study, the delay was proportional to the dosage rate.

PAGE 125

125 Higher dosage rates of ADVA 600 Type F superplasticizer can result in an increase in both modulus and compressive strength when compared t o the same mix with the same water to cement ratio but no admixture.

PAGE 126

126 CHAPTER 10 RECOMMENDATIONS FOR FUTURE RESEARCH The conclusions drawn from the work performed in this research point to many potential avenues for future research, both to improve the be havior of the model as well as to explore different material inputs and outputs. The empirical nature of the compressive strength model built into the VCCTL for the prediction of strength based on elastic modulus does not perform well for Florida limeston e. Generalized empirical predictions for strength of concrete are inherently limited due to the variability of the materials used in its production. A more fundamental approach to the simulation of concrete strength should be investigated that considers th e properties of the aggregate cement interface zone. Detailed characterization of the elastic properties of Florida limestone should be performed, with multiple measurements for multiple different sources. A database of properties such as density, absorpt ion, desorption, reactivity, and angularity for use with the VCCTL and other projects should be created. The VCCTL supports the modeling of admixtures if the specific phase surface deactivation behavior of the admixture is known. Methods of obtaining this information, either through material testing or possibly from the admixture manufacturer, should be investigated. There are a number of materials that can be modeled in the VCCTL that were not considered for this research. There is support for both fly ash and blast furnace slag hydration in the VCCTL, though the accuracy of the model in this respect is largely unknown. The techniques required to characterize these materials are also more involved due to the glassy nature of their composition. The methods b y which these materials can be characterized and the accuracy with which they are simulated in the VCCTL should be explored.

PAGE 127

127 In addition to the simulation of modulus and strength, the VCCTL computes a number of outputs that were not considered in this rese arch. Methods to evaluate the accuracy of these outputs should be explored with a focus on the following: Chemical Shrinkage Total porosity Solution Conductivity The techniques to measure these properties may already exist, or they may need to be explored

PAGE 128

128 APPENDIX A MATHEMATICAL EQUATIONS REFERENCE FOR THE CALCULATION OF ELASTIC MODULI USING DIFFERENTIAL EFFECTIVE MEDIUM THEORY The equations used to create a single effective particle are as follows Where: Where G is the effective shear modulus The effective bulk modulus, K, is given by: The differential equations used to solve numerically for the composite modulus are as follows

PAGE 129

129 Where Where m denotes a property of the matrix, and p denotes a property of the particle

PAGE 130

130 APPENDIX B SEM FIELD RAW DATA Table B 1: Fields Acquired At RJ Lee Field C3S Area Perimeter C2S Area Perimeter C3A Area Perimeter C4AF Area Perimeter 1 0.58 0.31 0.23 0.51 0.10 0.15 0.08 0.04 2 0.58 0.31 0.23 0.50 0.10 0.14 0.08 0.04 3 0.50 0.33 0.26 0.41 0.07 0.15 0.18 0.11 4 0.60 0.36 0.23 0.46 0.01 0.01 0.16 0.17 5 0.43 0.31 0.36 0.50 0.08 0.09 0.14 0.10 6 0.47 0.22 0.33 0.52 0.13 0.19 0.07 0.07 7 0.41 0.25 0.29 0.39 0.16 0.29 0.15 0.06 8 0.51 0.36 0.27 0.37 0.08 0.12 0.14 0.15 9 0.46 0.33 0.33 0.45 0.06 0.12 0.15 0.11 10 0.52 0.50 0.29 0.31 0.10 0.13 0.08 0.06 11 0.58 0.32 0.22 0.45 0.05 0.05 0.15 0.18 12 0.45 0.29 0.31 0.41 0.09 0.20 0.15 0.09 Avg. 0.51 0.32 0.28 0.44 0.09 0.14 0.13 0.10 Table B 2:Fields Acquired At UF Field C3S Area Perimeter C2S Area Perimeter C3A Area Perimeter C4AF Area Perimeter 1 0.47 0.26 0.32 0.66 0.05 0.06 0.13 0.01 3 0.47 0.22 0.45 0.71 0.01 0.02 0.07 0.04 5 0.41 0.16 0.34 0.64 0.10 0.11 0.17 0.08 6 0.57 0.34 0.27 0.38 0.11 0.19 0.03 0.06 9 0.52 0.51 0.28 0.31 0.02 0.01 0.18 0.16 Avg. 0.49 0.31 0.32 0.54 0.06 0.07 0.11 0.07

PAGE 131

131 LIST OF REFERENCES ASTM. (2002). Concrete in Compression. ASTM. (2007). C 192 Standard Practice for Making and Curing Concrete Test Specimens in the Laboratory. ASTM. (2008). C 403 Standard Test Method for Time of Setting of Concrete Mixtures by Penetration Resistance. ASTM. (2009). C 143 Standard Test Method for Slump of Hydrau lic Cement Concrete. ASTM. ASTM. (2009). C 39 Standard Test Method for Compressive Strength of Cylindrical Concrete Specimens. ASTM. (2009). C1702. Standard Test Method for Measurement of Heat of Hydration of Hydraulic Cementitious Materials Using Isothe rmal Conduction Calorimetry ASTM. (2009). Standard Test Method for Air Content of Freshly Mixed Concrete by the Volumetric Method. Bentz, D. P. (2005). Quantitative Comparison of real and CEMHYD3d model microstructures using correlation functions. Cement and Concrete Research 259 263. Bentz, D. P. (2006). Modeling the influence of limestone filler on cement hydration using CEMHYD3D. Cement and Concrete Composites 124 129. Bentz, D. P., Feng, X., Haecker, C. J., & Stutzman, P. E. (2000). Analysis of CCRL Proficiency Cements 135 and 136 Using CEMHYD3D. Gaithersburg: U. S. Department of Commerce. Bentz, D. P., Haecker, C., Feng, X., & Stutzman, P. (2002, September 23 27). Prediction of Cement Physical Properties by Virtual Testing. Process Technology of Cem ent Manufacturing pp. 53 63. Bullard, J. W. (n.d.). Curing of Concrete: Spatial and Temporal Randomness Over Multiple Length Scales. Bullard, J. W., & Garboczi, E. J. (2006). A model investigation of the influence of particle shape on Portand cement hydra tion. Cement and Concrete Research 1007 1015. Bullard, J., Garboczi, E., & Stutzman, P. (2013, June 13). VCCTL Software: Overview and Opportunities. Edmeades, R. M., & Hewlett, P. C. (1998). Cement Admixtures. In Lea's Chemistry of Cement and Concrete (pp 841 905). Elsevier.

PAGE 132

132 Elwell, D. J., & Fu, G. (1995). Comppression Testing of Concrete: Cylinders vs. Cubes. New York: Transportation Research and Development Bureau. Ferraro, C. C. (2009). Determination of Test Methods for the Prediction of the Behavior o f Mass Concrete. University of Florida. Florida Department of Transportation. (2014). Portland Cement Concrete. In Standard Specifications for Road and Bridge Construction (pp. 308 328). Garboczi, E. J. (2000). The past, present and future of computational materials science of concrete. Materials Science of Concrete Workshop Garboczi, E. J., & Bentz, D. P. (1998). The Microstructure of Portland Cement based Materials: Computer Simulation and Percolation Theory. Computational and Mathematical Models of Micr ostructural Evolution (pp. 89 100). San Francisco: Materials Research Society. Garboczi, E. J., & Day, A. R. (1995). An algorithm for computing the effective linear elastic properties of heterogeneous materials: Three dimensional results for composites wit h equal phase ratios. Journal of the Mechanical Physics of Solids 1349 1362. Garboczi, E. J., Bullard, J. W., & Bentz, D. P. (2004). Virtual Testing of Cement and Concrete USA 2004. Concrete International 33 37. Garboczi, E. J., Thorpe, M. F., & Day, A R. (1991). Universal conductivity curves for a plane containing random holes. Pysical Review 6473 6482. Garboczi, E., & Berryman, J. (2001). Elastic modului of a material containing composite inclusions: effective medium theory and finite element comput ations. Mechanics of Materials 455 470. Garboczi, E., Bullard, J., & Martys, N. (2010). The Virtual Cement and Concrete Testing Laboratory: Application to Sustainability. Concrete Sustainability Conference. Tempe, AZ: National Ready Mixed Concrete Association. Goldstein, J., Newberry, D. E., Joy, D. C., Lyman, C. E., Echlin, P., Lifshin, E., . Michael, J. (2007). Scanning Electron Microscopy and X ray Microanalysis. Springer. Hamlin N. Jennings, S. K. (1 986). Simulation of Microstructure Development During the Hydration of of a Cement Compound. Journal of the American Ceramic Society 790 799. Jones, R., & Kaplan, M. F. (1957). The effects of coarse aggregate on the mode of failure of concrete in compress ion and flexure. Magazine of Concrete Research 89 94. Mamlouk, M. C., & Zaniwski, J. P. (2010). Materials fo Civil and Construction Engineers. Prentice Hall.

PAGE 133

133 Maso, J. C. (1996). Influence of the interfacial transition zone on composite mechanical peoperties. London: E & FN Spon. McClellan, G. H., Eades, J. L., & Fountain, K. B. (1993). Petrographic Characteristics of Florida Limestone Aggregates used in Concrete Phase II. Final Report. Washington, D.C.: Transportation Research Board. Mindness, S. & Young, J. F. (1981). Concrete. Englewood Cliffs: Prentice Hall, Inc. Neville, A. M. (2011). Properties of Concrete. Trans Atlantic Publications, Inc. Peyrot, I. C. (2006). Development and Validation of a 3D Computational Tool to describe Damage and Fra ctue due to Alkali Silica Reaction in Concrete Structures. Russo, V. (2012, December). Towards a modelization of the effect of superplasticizers on early age mechanical strengths with VCCTL. Gainesville. Sahachaiyunta, P. (2012). Virtual Testing in a Ceme nt Plant. Concrete International 33 39. Valentini, L., Parisatto, M., Russo, V., Ferrari, G., Bullard, J. W., Angel, R. J., . Artoli, G. (2013). Prediction of the poperties of industrial cement mortars by computer modeling (preprint). Science and Tech nology of of Advanced Materials (Submitted) Waseda, Y., Matsubara, E., & Shinoda, K. (2011). X ray Diffraction Crystallography: Introduction, Examples and Sovled Problems. Springer. Wittke, J. H. (2008). Microprobe SEM. Retrieved from http://www4.nau.edu/ microanalysis/Microprobe SEM/Signals.html Wittmann, F. H., Roelfstra, P. E., & Sadouki, H. (1984 1985). Simulation and Analysis of Composite Structures. Materials Science and Engineering 239 248. X. Feng; NIST; Northwestern University. (2004). Expanding a tool for predicting chloride diffusivity in concrete so it can be used by manufacturers to evaluate the durability of concrete made with blended cements. U. S. Department of Commerce. Zeinkiewicz, O. C., Taylor, R. L., & Zhu, J. Z. (2005). The Finite El ement Method: Its Basis and Fundamentals. Butterworth Heinemann.

PAGE 134

134 BIOGRAPHICAL SKETCH Benjamin Watts was born to Lori Reeves and Daniel Watts in January of 1989 in Seattle, Washington. He attended primary school in Florida, and began studying at the University of Florida in the fall of 2007. There he r il e ngineering, after which he was grant ed the opportunity to pursue a m d egree.