This report presents a study of three typical tailings samples as potential cemented backfill in underground mines. The testing series was unique in that the pulp densities of the samples were all above 75 pct solids. Test results included dry density; slump; percent settling after 28 days of curing; tensile strength after 28, 120, and 180 days of curing; and unconfined compressive strengths after 7, 28, 120, and 180 days of curing. The physical properties of the various test mixtures were further analyzed using linear and nonlinear statistical methods to produce correlations and mathematical equations. Physical properties were used to determine the influence of mix additives and as input for numerical modeling studies of backfill. The mathematical relations were used as a predictive tool in determining the suitability of various materials as backfill.

Conventional room-and-pillar mining has been commonly used in the United States. However, the domestic mining industry has been hardpressed to maximize mineral productivity in order to compete with foreign suppliers. As a consequence, U.S. mines no longer have the luxury of leaving ore-rich pillars as ground support, and reserves tied up in highly fractured material cannot be left behind. Existing mines are also encountering greater ground stresses as mining progresses deeper, causing the openings to squeeze inward dramatically or suddenly burst. In certain regions of the country and particularly near urban areas, ground subsidence poses safety and environmental hazards.

Backfilling stopes allows removal of pillars in addition to controlling ground subsidence. The fill also acts as a medium with established engineering properties and predictable behavior. These advantages can allow mines to maximize their ore reserves.

Backfill has been extensively used worldwide. The Bureau was involved as far back as 1964 in defining the properties of hydraulically placed backfill. From 1961 to 1970, Canadian researchers tested a multitude of mixes utilizing portland cement and mine tailings. These tests all used the common mode of hydraulic transport of materials that were typically <70 pct solids. At 45 pct water content, the backfill needed to be designed for high permeability where the excess water was drained and pumped out of the mine. Bleeding of the cement and aggregate fines through the drainage water was a common problem, resulting in greatly varied in-place strength.

Today, pumps and pneumatic blowers are capable of handling a mine’s rugged environmental requirements while meeting a 100-st/h operating speed. These new pumps and pneumatic stowers may make it favorable to transport >80 pct solids, total tailings material from the mill to the stope for use as backfill. This capability no longer limits the mix matrix to 70 pct solids or to the inclusion of only the sands fraction of mill tailings. The resulting decrease in water improves the strength, homogeneity, and curing time of the material, and makes lean, cemented total tailings backfill an attractive option.

This report summarizes laboratory work done by the Bureau to define the strength characteristics of lean, cemented backfill using total tailings as aggregate, and varying the cement and other additives as well as the water content. The mix matrix used simulated the higher pulp densities capable of being transported and placed by large concrete pumps, physical stowing equipment, pneumatic blowers, or gravity free fall.

Test Procedure

Mix Composition

The mill total tailings used as the basic aggregate in this test series came from three underground metal mines: Tailings A from a deep silver mine in Idaho, tailings B from a lead-zinc mine in Missouri, and tailings C from a copper-silver mine in Montana. Grain-size gradation curves are shown in figure 1. The fines content of the total tailings was retained to minimize the void ratio. This practice has been documented as improving strengths and decreasing fill consolidation. Mix matrices are summarized in appendix A.

Commercially available Type I and II portland cement and tap water were used in all mixes in the test series. The following additives were incorporated in the test mixes to determine their influence on some of the physical properties of the tailings.

Fly Ash—Various mixtures of commercially available ASTM Class F fly ash were added to the tailings to determine whether the pozzolanic influence would be sufficient to decrease the required amount of cement and still maintain the unconfined compressive strength.

Pit-Run and Ground Smelter Slag.-The cementing influence of the smelter slag was determined by Construction Technology Laboratories of Skokie, IL. The chemical analysis is shown in table 1. Since the hydraulicity, or the ability of the slag to react with water, is believed to increase when the slag is ground very fine, the tests included different gradations of ground slag. Slag samples of 400, 500, and 600 m²/kg as determined by the Blaine test were mixed with water and showed no unconfined compressive strengths through 28 days of curing time because the material remained in the original slurry state. Grain-size analyses of the pit-run and ground smelter slag are shown in figure 2.

Oil Shale Retorted Waste.-Because previous oil shale research had documented the cementing properties of certain retorted wastes, oil shale retorted waste was used as an additive to determine if its cementing properties could be used in the backfill. The grain-size gradation curve of the retorted waste is given in figure 2.

Kiln Dust.-A locally available source of kiln dust was used in a few mixes to determine its cementitious effects.

Superplasticizer.-An ASTM Category B superplasticizer was added to a few mixes to determine its ability to decrease the amount of water necessary to initiate cementing action and maintain pumpability. Ten times the

manufacturer’s recommended dosage rate for concrete was needed before a measurable increase in slump was seen. Since results of the, superplasticizer tests were not promising, no further tests were attempted.

Note.-2.24 pct is accounted for by other concentrations or combustion losses.

Mix Handling

The mixing procedure for the test series included use of a portable cement mixer. After the oven-dried total tailings material and any additives were mixed for a minimum of 2 min and visually checked for homogeneity, a slump measurement was taken. Eight samples were taken from each test mix, packed into standard 3- by 6-in, waxed cardboard cylinders, and cured in a fog room. The slurry density was taken at the time of mixing, and the 28-day wet density was measured after 28 days of curing. Gang molds were cast using the various mixes for tailings B and C to obtain samples for determining tensile strength. These briquets were also cured in a fog room.

Laboratory Testing

The test series included unconfined compressive strength determinations after 7, 28, 120, and 180 days of moist curing. Each strength test was run on a duplicate cylinder sample and the two strength readings were averaged to minimize errors. Eight tailings A cylinder samples were tested: two each for 7-, 28-, and 120-day cured, unconfined compressive strength tests; one for the 120- day cured, confined compressive strength tests; and one for determination of the dry density. Eight tailings B and C samples were tested: two each for 7-, 28-, 120-, and 180- day cured, unconfined compressive strength tests. In addition, the test results of three briquet specimens were averaged to determine 28- and 120-day cured tensile strengths for tailings B and C.

Initial mixes of the total mill tailings were cast without benefit of binder (cement, fly ash, etc.). These samples remained in a slurry state and did not achieve a compressive strength. In addition, the saturated environment of the fog room prevented any evaporation from taking place.

Appendix A summarizes the mix proportions along with the various additives, the types of tests conducted, and the test results. Cement, fly ash, pit-run smelter slag, ground smelter slag, kiln dust, and oil shale retorted waste were measured as a percentage of the total tailings aggregate (dry weight of fly ash plus pit-run smelter slag plus ground smelter slag plus kiln dust divided by dry weight of total tailings times 100). The water-to-cement ratio was calculated as a proportion of the weight of the water to the weight of the cement. The water-to-binder ratio was used to determine if the additives influenced the cementing properties of the mix and was calculated as the proportion of the weight of the water to the combined weight of the cement, fly ash, kiln dust, and oil shale retorted waste. Because the slag was known as a nonhydraulic additive, it was not included as a “binder.”

The slurry density of the mixes was determined by dividing the weight of the solids by the weight of the solids and water. Slump measurements were taken to determine possible pumpability of the various mixes and was measured in inches. The tensile and unconfined compressive strengths (measured in pounds per square inch) were averaged through use of replicate testing.

Test Analysis

Laboratory Tests

Category B superplasticizer did not seem to have an impact on reducing the water content and increasing the workability of the mixes. This may have resulted from the nature of the superplasticizer used in the tests. The particular superplasticizer used was Mighty 150, a sulphonated naphthalene formaldehyde condensate. This type of superplasticizer reacts by increasing the charge of the cement particle, thereby repelling the individual cement particles from each other and resulting in better dispersion throughout the mix. It also decreases the surface tension of the water, making it “wetter.” In these lean mixes (<15 pct cement content), the influence of Category B superplasticizers would be diminished.

Statistical Analysis

At the beginning of the test series, tailings A results were statistically analyzed to determine if any meaningful relationships existed among the data. Ninety-five sample pairs were cast through the course of testing and included various additives such as pit-run smelter slag, ground smelter slag, and fly ash. Four candidate predicted variables (variables to be predicted from mixture information) were measured: average unconfined compressive strength at 7-, 28-, and 120-day curing increments and the resulting slump value. There were five predictor variables: pit-run smelter slag, ground smelter slag, fly ash, cement, and water-to-cement ratio. The predictor variables were mixed in varying proportions to test for fill properties of the tailings.

The Minitab statistical computer program was used for most of the analytic work and the primary statistical algorithm was a multivariate linear model. With 14 variables involved, it was necessary to perform a preanalysis of the variables that would sort out some of the more spurious prior to the multivariate model fitting. Therefore, pair-wise correlation coefficients between the variables involved were examined first. These values are summarized in table 2 and provide a quick method to determine which variables are most highly correlated.

The matrix in table 2 is a mixture of both predictor and predicted variables, as defined previously. An absolute value of 0.8 correlation coefficient was arbitrarily chosen to delineate significance (a 1.0 correlation coefficient is a perfect fit of the line to the data points). Using this criterion, 10 pair-wise relationships were deemed significant; however, three of these correlations were between the dependent variables themselves, i.e., 28-day average unconfined compressive strength (28DCOMP) versus 120-day average unconfined compressive strength (120DCOMP), with a correlation coefficient of 0.944.

Various combinations of variables were further analyzed by least squares fitting a three-dimensional (3-D) hyperplane, which yielded the following equation:

7DCOMP = -73.6 + 32.4 CEMENT – 1.40 W/C
where
7DCOMP = 7-day unconfined compressive strength, psi,
CEMENT = cement content, pct,
and W/C = water-to-cement ratio.

The analysis provided an R² value of 0.876. The various goodness-of-fit parameters r, R², and I are discussed in appendix B. The results of the linear regression analysis are listed in appendix C.

Extreme scatter in some of these data is apparent in the itemized predicted variable (predicted Y-value, Fit column) versus the actual data (7DCOMP column) in appendix C. To illustrate, observation 30 of appendix C lists an actual 7-day observed strength of 160 psi. However, the predicted value using the regression equation produced a result of 343.35 psi. For this reason, two other modeling schemes were investigated: a multivariate, linear stepwise regression model and a univariate, nonlinear exponential model. The differences between these models are described in reference 15.

To determine a best multivariate linear model, stepwise regression was applied to the tailings A data. Briefly, this is a procedure that picks the predictor variables one at a time in order of relative importance. This approach has two advantages to the user: it produces a linear model to represent the data, and in so doing, it searches for the most important subset of dependent variables that will do the job. The Bureau’s stepwise code has an additional advantage in that it allows the creation of variables that are derived from the original predictor variable set. For example, cement and fly ash content were predictor variables. Terms involving cement or fly ash squared, cubed, multiplied, raised to powers, etc., can be easily inserted in the model. There is one important aspect, however, which must be kept in mind when using this model. In forming the regression, the user is always fitting an additive model of the terms of interest.

The stepwise procedure was applied individually to each of the predicted variables involved: 7-, 28-, and 120-day cured, unconfined compressive strengths and the slump variable.

Mathematical representation of the stepwise regression model is given by

Y = P1 + P2Z1 + P3Z2 + P4Z3 + . . . + P1Zk,
where
Y = predicted variable (here, 7-day unconfined strength),
P1,P2,……………….Pi = constants found by the stepwise process,
and
Z1,Z2,…………..Zk = selected predictor variables (cement, fly ash, etc.) chosen one at a time in order of importance.

It was necessary to use four predictor variables (cement, pit-run smelter slag, water-to-cement, and fly ash), of which only cement and pit-run smelter slag were deemed statistically significant, to produce an equation predicting the 7-day unconfined compressive strength with an R² value of 0.887 (table 3).

Another modeling attempt was made using a two-dimensional model with the predicted and predictor variables fitted by an exponential curve. Figure 3 illustrates the data by plotting the 7-day unconfined compressive strengths to the water-to-cement ratio. This curve-fit procedure resulted in the following equation:

Table 3 tabulates the various results of the three statistical methods used to determine goodness-of-fit for the tailings A test data as applied to predicting the 7-day unconfined compressive strengths. The 3-D hyperplane produces a correlation coefficient of 0.876. In the multivariate, linear stepwise regression model, the anticipated 7-day unconfined compressive strengths fit the observed compressive strengths of each test specimen with a correlation coefficient of 0.887. The predictor variables, listed by order of importance to determine the 7-day compressive strength, are cement and pit-run smelter slag. The nonlinear, exponential model produces an index of determination (see appendix B for definition) of 0.889, which is quite promising since it is based on only one input variable, the water-to-cement ratio.

After it was determined that the exponential model would best fit the data curves of the unconfined compressive strengths versus the water-to-cement ratio, the data from tailings A, B, and C were analyzed as a group for comparison. Plots of the compressive strengths versus water-to-cement ratios for the total data base are presented in figure 4. The mathematical representation of the curves along with their respective indices of determination are given in appendix D.

Further analysis of the tailings A, B, and C data included exponential curve fitting of the compressive strengths to water-to-cement ratios for the mixes grouped by tailings type and then by tailings type not containing any additives (pit-run smelter slag, ground smelter slag, fly ash, kiln dust, and oil shale retorted waste) (figs. 5-6). The mathematical representation of the curves along with their respective indices of determination are also given in appendix D.

As can be seen in appendix D, the goodness-of-fit increases as the data base becomes increasingly selective. For instance, the total data base index of determination, I, for 7-day compressive strength is 0.796. For the data base containing only tailings A, I is 0.889; and for the tailings A data base not containing any additives (pit-run smelter slag, fly ash, etc.), I is 0.982.

Discussion of Results

The results indicated that the addition of oil shale retorted waste without the benefit of cement produced compressive strengths on the order of 100 psi in 28 days. The cementing properties of the retorted waste were greater for the finer particles of tailings C. The addition of fly ash improved the compressive strength of the total tailings aggregate. As the tailings grain size fraction greater than 200 sieve increased, the influence of the fly ash decreased. The 28-day compressive strength of tailings A was increased by 25 pct, tailings B by 48 pct, and tailings C by 98 pct over the compressive strengths gained by the use of cement alone.

As the grain size of the tailings fraction greater than 200 sieve decreased, the compressive strengths also decreased for the various curing periods. The 7-day compressive strength for tailings A with 6 pct cement and a water-to-cement ratio of 4.5:1 was 118 psi; for tailings B it was 107 psi; and for tailings C it was 65 psi.

The linear relationship (based on least squares fitting) between the 7-day compressive strengths and those of 28, 120-, and 180-day compressive strengths is presented in figure 7. In each case, the strength gained between each pair of relationships was greater as the grain size of the tailings material increased. This was just the opposite for the relationship between the 7-day compressive strength and the 28-day tensile strength (fig. 7). There was no significant difference in strength gain between the 28-, 120-, and 180-day compressive strengths and the 7-day compressive strength because of grain-size differences (fig. 8). However, grain size differences caused a marked difference between the 28-day tensile strength and the 7-day compressive strength (fig. 8). The finer-grained tailings C developed a higher tensile strength when compared to the 7-day compressive strength.

The ratios between compressive strength to tensile strength for the various days of curing ranged from 4.4 for

the total tailings not containing additives to 4.8 for the total data base.

The goodness-of-fit for calculating the compressive strengths using the water-to-cement ratio and the exponential formula is

Y = Ae-BX,
where Y = compressive strength, psi,
X = water-to-cement ratio,

and A and B are constants.

This goodness-of-fit progressively improves as the sample groupings become more restricted. Appendix D summarizes the indices of determination for the various groupings.

Summary

This series of backfill material testing was initiated to determine what engineering properties could be expected from a variety of mill total tailings. The incorporation of additives was meant to define the extent of increased strength or workability of the resultant mix. None of the tailings tended to be self-cementing. However, as cement contents were increased, compressive strengths increased. The compressive strengths of the fly ash-and-cement combination increased after 28 days of curing as compared to the strength of cement alone after 28 days. The addition of pit-run smelter slag, which incorporated coarser particles into the mix, seemed to increase compressive strength, but the slag alone was not cementitious. In some cases, such as those where oil shale retorted waste was added to the tailings, a full range of mixes was not attempted since the problem was merely to determine whether or not the retorted waste was a detriment to the mix, thereby indicat¬ing possible uses of oil shale waste.

Further research will test the relationships found during this investigative test series. The effects of chemical additives such as superplasticizers, high-early-strength cements, water-reducing agents, and kiln dust will be examined. With further refinement, an accurate predictive tool will be developed that will assist the industry in analyzing the suit¬ability and stability of dewatered, total-tailings backfill.