Particle Shape, Type and Amount of Fines, and Moisture Affecting Resilient Modulus Behavior of Unbound Aggregates Debakanta Mishra 1, Erol Tutumluer 2, M. ASCE, Yuanjie Xiao 3 1 Graduate Research Assistant, Department of Civil and Environmental Engineering, University of Illinois Urbana-Champaign, Urbana, IL, 61801, USA, dmishra2@illinois.edu 2 Professor, Department of Civil and Environmental Engineering, University of Illinois Urbana- Champaign, Urbana, IL, 61801, USA, tutumlue@illinois.edu 3 Graduate Research Assistant, Department of Civil and Environmental Engineering, University of Illinois Urbana-Champaign, Urbana, IL, 61801, USA, xiao8@illinois.edu ABSTRACT: This paper presents the application of Analysis of Variance (ANOVA) approach to study effects of different aggregate properties on the resilient modulus behavior of unbound aggregates. In a recent research project at the University of Illinois, stress-dependent resilient modulus characterization models, in the form proposed for use by the Mechanistic Empirical Pavement Design Guide (MEPDG), were established for three different aggregate types (dolomite, limestone, and uncrushed gravel) using a statistically well-formulated laboratory test matrix. Such a consistent database enabled the evaluation of the model parameter dependence on different aggregate properties, such as particle shape, type and amount of fines, and moisture condition, using the Analysis of Variance (ANOVA) approach. From the statistical analyses, aggregate particle type or angularity was found to have a significant effect on the resilient moduli. At low fines content, the type of fines and moisture state in relation to optimum moisture condition were found to be somewhat insignificant as far as the regression parameter values are concerned. The effect of amount of fines became quite significant especially when the amount of fines was increased. INRODUCTION Recent research at the University of Illinois focused on evaluating effects of aggregate type and quality on aggregate cover thickness requirements for constructing pavement platforms on soft subgrade (Tutumluer et al., 2009; Mishra et al., 2010). The overall objective in the first laboratory phase of the study was to develop a statistically well-formulated test matrix to quantify any individual and/or combined effects of different aggregate properties on the unbound aggregate layer performance as indicated by the strength, stiffness, and deformation behavior. Three aggregate 279
280 GEOTECHNICAL SPECIAL PUBLICATION NO. 203 materials, i.e., dolomite, limestone and uncrushed gravel, commonly used in Illinois for subgrade replacement and subbase were obtained to determine their particle shape and angularity quantified by an imaging based measurement system, type and amount of fines (passing No. 200 sieve size or less than 0.075 mm), and moisture state in comparison to optimum moisture content. Evaluating the effects of these properties on aggregate performance would enable researchers and practitioners to establish corresponding property ranges so as to ensure adequate performance of the pavement layers comprised of these aggregates. For studying the effect of fines on aggregate behavior, it was decided to test the laboratory specimens at four different fines contents. These fines contents were established, to range from very low to very high, as allowed in the field by state specifications. Accordingly, the target fines contents were established to be 4%, 8%, 12% and 16% material passing No. 200 sieve size or particles smaller than 0.075 mm. All aggregate gradations were engineered to have a consistent particle size distribution across all samples and therefore eliminate unwanted variation in aggregate performance due to gradation differences. Figure 1 shows the engineered gradations used to blend samples for testing. 100 100 10 1 0.1 0.01 Cumulative Percent Passing 80 60 40 20 0% Fines 4% Fines 8% Fines 12% Fines 16% Fines 0 100 10 1 0.1 0.01 Particle Size (mm) FIG.1. Engineered aggregate gradations for different fines contents To study the effect of type of fines on aggregate behavior, two different types of fines were used. One was non-plastic in nature such as mineral filler type (with plasticity index PI = 0), and the other was plastic such as cohesive fine-grained soil type (PI in the range of 10-12). The effect of moisture state on aggregate performance was studied by testing the blended aggregate specimens at three different moisture contents: (1) optimum moisture content (OMC or w opt ), (2) 90% of w opt, and (3) 110%
GEOTECHNICAL SPECIAL PUBLICATION NO. 203 281 of w opt, where the w opt was established through the standard Proctor (AASHTO T-99) test for each aggregate gradation. Therefore, the laboratory test matrix ended up being a 4x2x3 factorial (4 different fines contents, 2 different types of fines, 3 different moisture contents) for each aggregate type. Although a comprehensive database of aggregate performance indicators was generated as a result of this research project, the current paper focuses on the resilient modulus test results only. Therefore, this paper is aimed to identify the relative significances of these aggregate properties affecting the resilient modulus behavior through the Analysis of Variance (ANOVA) approach. RESILIENT MODULUS CHARACTERIZATION Starting with the 1986 AASHTO Pavement Design Guide in the U.S., resilient modulus has been used to characterize the stiffness behavior of subgrade soils and granular base/subbase layers subjected to repeated traffic loading. With the routine use of resilient modulus as a primary input into mechanistic-empirical pavement design procedures, many state highway agencies are now making an effort to establish the appropriate resilient modulus inputs for granular materials and subgrade soils. As a measure of a material s ability to withstand loading, resilient modulus plays an integral part in the thickness design and long term performance of a pavement. Among several different characterization models proposed for expressing the stress dependent resilient modulus values of unbound aggregate materials, the following form has been adopted by the recent NCHRP 1-37A project deliverable Mechanistic Empirical Pavement Design Guide (MEPDG) software (http://www.trb.org/mepdg/): 1 2 3 K2 K3 θ τ oct MR = K1 pa + 1 pa pa where M R = Resilient Modulus p = Atmospheric Pressure Normalization Factor a θ = Bulk Stress (First Stress Invariant) τ oct = Octahedral Shear Stress K, K, K : Regression Parameter Estimates The M R test data from the research project database was fitted with the above model and values of the model parameters, K 1, K 2, and K 3, were estimated. The test results showed very good fit with the model and the R 2 (coefficient of determination) values obtained for each of the individual test samples were significantly high, above 0.96. To evaluate the dependence of the M R model parameters with different aggregate properties, an Analysis of Variance (ANOVA) approach was adopted to conduct regression of the model parameters, K 1, K 2, and K 3, individually against each of the four classification variables, i.e., aggregate angularity, amount of fines, type of fines, and moisture condition. The statistical SAS TM software package was used for carrying out all statistical computations. When the actual nonlinear and anisotropic nature of
282 GEOTECHNICAL SPECIAL PUBLICATION NO. 203 the resilient modulus behavior is concerned, results for the constant term K 1 were found to be primarily governing the moduli behavior at a certain stress state (Tutumluer, 2009). Accordingly, these aggregate property effects on the K 1 model parameter are presented in this paper for brevity although detailed results for the effects on K 2 and K 3 have been thoroughly investigated as part of the ongoing efforts at the University of Illinois. The individual aggregate properties influencing K 1 values were found to be interacting with each other. In other words, the effect of amount of fines on K 1 changed depending on the moisture condition as well as the type of fines. Therefore, when studying the individual effects of these parameters on K 1, the aim was to evaluate the effect of each individual property averaged over all possible combinations of the other properties. Another research effort is currently underway to consider the combined effects of all these properties on resilient modulus model parameters and develop prediction equations for the estimates of the K values. Moreover, future work will also focus on conducting full factorial analyses of the data to evaluate the overall effects of different property combinations. EFFECT OF PARTICLE SHAPE Aggregate particle shape, texture and angularity have been found to affect performances of unbound and bound layers in pavement structures (Allen, 1973; Barksdale and Itani, 1989; Rao et al., 2002). Unbound aggregate layers having crushed particles have consistently performed superior compared to those with uncrushed particles. Allen (1973) and Barksdale and Itani (1989) investigated the effects of surface characteristics of unbound aggregates and found that angular materials resisted permanent deformation better than rounded particles because of improved particle interlock and higher angle of shear resistance between particles. Barksdale and Itani (1989) also concluded that blade shaped crushed particles were slightly more susceptible to rutting than other types of crushed aggregate and that cube-shaped, rounded river gravel with smooth surfaces was more susceptible than crushed aggregates. More recently, Rao et al. (2002) studied the impact of imaging based aggregate angularity index variations on the friction angle of different aggregate types and reported an improvement in aggregate performance when the percentage of crushed particles increased. The three aggregate types studied were first tested for particle shape, texture and angularity characteristics. The use of a validated image analysis system, the University of Illinois Aggregate Image Analyzer (UIAIA), was pursued to quantify aggregate shape (flatness and elongation), angularity and surface texture characteristics. The indices computed were: Flat and Elongated (F&E) ratio (Rao et al., 2001), Angularity Index (AI) (Rao et al., 2002), and Surface Texture (ST) index (Rao et al., 2003). The average AI values obtained for the three aggregate types were: 330 for gravel, 428 for dolomite, and 481 for limestone, respectively. The variance in K 1 with AI was analyzed, and the results are presented in Figure 2 which shows that the p-value corresponding to AI is <0.0001. Taking a significance level (or type 1 error rate α) of 0.05, it can be concluded that the mean K 1 values are not the same across all the three aggregate types. In other words, AI has some effect
GEOTECHNICAL SPECIAL PUBLICATION NO. 203 283 on the K 1 values. Although the ANOVA results proved that the K 1 values were affected by aggregate particle angularity, they do not give direct information about whether all three aggregate types were different from each other, or only one aggregate type had different K 1 values than the other two. This question was answered by running two linear contrasts (each with an α level of 0.05) to compare the K 1 value distribution for dolomite with gravel, as well as limestone. The SAS TM output from the linear contrasts is presented in Figure 3, which clearly shows a significant difference between the K 1 values of uncrushed gravel and crushed dolomite (p <0.0001). However, the K 1 values of dolomite and limestone did not have a statistically significant difference among them (p = 0.0940). This required grouping the commonly used aggregates into two categories crushed and uncrushed to account for the particle shape. Accordingly, crushed particles resulted in significantly higher resilient moduli. EFFECT OF FINES TYPE The plastic fines often found in uncrushed aggregate gradations were highlighted in a recent Illinois Department of Transportation (IDOT) field study, Experimental Feature IL 03-01, to considerably influence the performances of aggregate cover layers in subgrade applications (IDOT SSM, 2005). From the SAS ANOVA analysis output, Figure 4 indicates that the p-value corresponding to type of fines is not significant at a significance level (α) of 0.05. This is indeed interesting that the type of fines does not have any significant effect on the K 1 model parameter, and the distributions are not statistically different. Plastic fines may in fact cause stress softening behavior to be potentially dictated by the K 3 exponent of the deviator stress. FIG.2. ANOVA output for the effect of AI on K 1 FIG.3. Orthogonal linear contrasts for comparing the effect of AI on K 1
284 GEOTECHNICAL SPECIAL PUBLICATION NO. 203 FIG.4. ANOVA output for the effect of fines type on K 1 EFFECT OF AMOUNT OF FINES The samples were blended targeting fines contents (percent passing No. 200 sieve or particles smaller than 0.075 mm) of 4%, 8%, 12%, and 16%. Although the actual fines percentages achieved after blending were not exactly equal to these targeted values, still they were always close to the target values. An ANOVA analysis was conducted to study if the aggregate M R model K 1 parameters were affected by the fines content in the sample. It should be noted that the fines contents were just treated as aggregate property variables and no correlations were developed to predict K 1 values from the fines percentages. Developing such correlations to predict M R model parameters from different aggregate properties is currently underway and the results will be published in the future. From the ANOVA output presented in Figure 5, it can clearly be seen, that an overall effect of the amount of fines on K 1 is not significant (p >> α =0.05). However, to further analyze the effect of fines content on K 1, linear contrasts were run to compare the values for 4% fines content against each of the other three. Results from the linear contrasts are presented in Figure 6. From the linear contrasts, as the amount of fines is increased from 4% to 8%, the difference in K 1 values becomes highly insignificant (shown by a p-value of 0.4055). However, as the fines content is increased gradually, the difference in K 1 values becomes more and more significant (p-value goes to 0.1512 and then to 0.0825). Although, the K 1 values for 16% fines is still not significantly different from that for 4% fines for an α level of 0.05, the difference in K 1 values becomes more and more significant as the fines contents change from 4% to 16%. Thus it may be possible to group aggregates into two broad categories, of low fines and high fines as far as resilient modulus characterization is concerned.
GEOTECHNICAL SPECIAL PUBLICATION NO. 203 285 FIG.5. ANOVA output for the effect of amount of fines on K 1 FIG.6. Linear contrasts for comparing the different amounts of fines on K 1 EFFECT OF MOISTURE CONDITION The samples were tested at optimum, dry of optimum, and wet of optimum conditions to study the effect of moisture content on the permanent deformation and shear strength values reported elsewhere (Tutumluer et al. 2009, Mishra et al., 2010). The effects of moisture condition on the K 1 parameter were further analyzed. The ANOVA results showed that the overall effect of moisture condition (dry vs optimum, or wet) is highly insignificant (p-value of 0.7710). Further analyses through linear contrasts were also conducted to check the individual differences between dry of optimum to optimum, and also optimum to wet of optimum (see Figure 7). However, the results clearly show that moisture condition does not contribute significantly as far as the variation in the values of K 1 is concerned. Although some specimens in the sample data may differ significantly, overall, the K 1 value distributions corresponding to the three moisture conditions were not statistically different. As a result, the moisture or suction effects may need to be taken into account by the bulk stress term and the K 2 exponent in the M R model.
286 GEOTECHNICAL SPECIAL PUBLICATION NO. 203 FIG.7. Linear contrasts for comparing the effect of moisture condition on K 1 CONCLUSIONS This paper presented a statistical analysis approach to identify the effects of four different aggregate properties, i.e., particle shape, type and amount of fines, and moisture condition, on the resilient modulus (M R ) model parameter estimates of unbound aggregate structural layers. A statistically well-formulated laboratory test matrix was used to develop the M R characterization model as a function of the aggregate properties. Analysis of Variance (ANOVA) analyses were conducted to study the individual effects of the four different classification variables. Aggregate angularity (crushed vs uncrushed) had a significant effect on the values of K 1, which is the constant term in the stress-dependent M R characterization model (K 2 and K 3 are the exponents of the bulk and deviator stresses, respectively). The K 1 values for uncrushed gravel were also significantly different from those of crushed dolomite. However, there was not much difference in the K 1 values for dolomite and limestone. This was expected since both materials comprised of crushed aggregates and had similar mineralogical compositions. Type of fines (plastic vs non-plastic) or moisture condition (dry of optimum, optimum, or wet of optimum) did not have significant effect on the K 1 model parameter. Plastic fines may in fact cause stress softening behavior to be potentially dictated by the K 3 exponent of the deviator stress whereas the moisture or suction effects may need to be taken into account by the bulk stress term and the K 2 exponent in the M R model. The overall effect of amount of fines on K 1 value was found to be insignificant. However, upon further analysis using linear contrasts, it was observed, that the difference between K 1 values of various fines percentages, i.e., 4% and 16%, becomes more and more significant, as the amount of fines increases. Further work is currently underway to develop correlations to predict the K parameters based on different aggregate properties. REFERENCES Allen, J. (1973). The Effect of Non-constant Lateral Pressures of the Resilient Response of Granular Materials. Ph.D. Thesis, University of Illinois at Urbana- Champaign, Urbana, IL. Barksdale, R.D. and Itani, S.Y. (1989). Influence of Aggregate Shape on Base Behavior, Transportation Research Record 1227, TRB, National Research Council, Washington, D.C., pp. 173-182 Rao, C., Tutumluer, E., and Stefanski, J.A. (2001). Flat and Elongated Ratios and Gradation of Coarse Aggregates Using a New Image Analyzer, ASTM Journal of Testing and Standard, Vol. 29, No. 5, 2001, 79-89.
GEOTECHNICAL SPECIAL PUBLICATION NO. 203 287 Rao, C., Pan, T., and Tutumluer, E. (2003). Determination of Coarse Aggregate Surface Texture Using Imaging Analysis, In Proceedings of the 16th ASCE Engineering Mechanics Conference, University of Washington, Seattle, WA. Rao, C., Tutumluer, E., and Kim, I.T. (2002). Quantification of Coarse Aggregate Angularity based on Image Analysis, Transportation Research Record 1787, National Research Council, Washington DC, pp. 17-124. Illinois Department of Transportation IDOT (2005). Subgrade Stability Manual, Bureau of Bridges and Structures, May 1, 2005, 27 pages. (http://www.dot.il.gov/bridges/pdf/subgrade stability manual.exe) Tutumluer, E. (2009) State of the Art: Anisotropic Characterization of Unbound Aggregate Layers in Flexible Pavements, In ASCE Geotechnical Special Publication No. 184, Pavements and Materials Modeling, Testing and Performance, ISBN: 978-0-7844-1008-0, pp. 1-16. Tutumluer, E., Mishra, D., and Butt, A.A. (2009). Characterization of Illinois Aggregates for Subgrade Replacement and Subbase, Final Report, ICT R27-1 Project, University of Illinois Urbana Champaign, Urbana, Illinois, 61801. Mishra, D., Tutumluer, E., and Butt, A.A. (2010). Quantifying Effects of Particle Shape and Type and Amount of Fines on Unbound Aggregate Performance through Controlled Gradation, Manuscript Accepted Presentation and Publication at the 89 th Annual Meeting of the Transportation Research Board, Washington, D.C., January 2010.