Evaluation and Modeling of Pressure Filtration of Coal Refuse Slurry

Open Access
- Author:
- Sankara Raman, Gireesh Subramaniam
- Graduate Program:
- Energy and Mineral Engineering
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- May 18, 2017
- Committee Members:
- Dr. Mark S. Klima, Dissertation Advisor/Co-Advisor
Dr. Mark S. Klima, Committee Chair/Co-Chair
Dr. William A. Groves, Committee Member
Dr. Jeremy M. Gernand, Committee Member
Dr. Herschel A. Elliott, Outside Member - Keywords:
- Coal Refuse Slurry
Pressure Filtration
Artificial Neural Network
Waste Management
Dewatering
Coal
Fine Coal
Mineral Processing - Abstract:
- Coal contributed to about 33% of total U.S electricity generation in the year 2015. Of the total coal produced in the U.S that year, about 45% was bituminous coal. Typically, this coal will require washing prior to utilization, which is carried out using density- and/or surface-based separation processes. After washing, the dewatered clean coal is sent to the power plant, while the coarse and fine refuse are often sent to impoundment areas. Every year, millions of gallons of coal refuse slurry is generated and stored in these impoundments, which potentially can lead to environmental problems such as acidic water, siltation of streams, land property devaluation and in extreme cases, dam failures. Refuse slurry could also get mixed with the local waterways by slow erosion of the berm surrounding abandoned ponds. Therefore, an alternative approach to slurry impoundments could reduce the overall environmental impact of coal processing. One method of reducing, if not eliminating, the need for slurry impoundments is to implement pressure filtration, a solid-liquid separation process, which provides a concentrated solid stream (cake) and liquid stream (filtrate) as products. Pressure filtration of coal refuse has the potential to improve water conservation, minimize environmental impact, reduce area requirements for disposal and provide a safer operation. This research is an attempt to evaluate, model, and scale-up pressure filtration to provide information that will help to better manage coal waste, thereby reducing the environmental impacts of coal mining and utilization. Evaluation with statistical experimental design and modeling with artificial neural network and linear regression were performed to study the pressure filtration of coal refuse slurry using bench and lab-scale filters for two refuse materials. Bench-scale testing was used to evaluate the effects of pressure, pH, solids concentration, temperature, fines fraction of solids, and filtration time. Results indicated an increase in cumulative filtrate flux, defined as the volume of filtrate per unit time per unit filter area, with increase in pressure and temperature while a decrease in flux was observed with increase in fines fraction, pH, filtration time, and solids concentration. Based on the experimental data, a 6-9-1 artificial neural network (ANN) model, with R2 values of 0.93 and 0.98 for the training and testing datasets, respectively, was developed to model filtrate flux as a function of pressure, slurry pH, feed solids concentration, fines fraction of solids in the slurry, filtration time, and temperature. The model was used as a tool for performing a novel method of sensitivity analysis on pressure filtration using Lek’s profile method. Results from the sensitivity analysis showed filtration time and pH to be the most significant variables influencing filtrate flux. Analysis of process scale-up in terms of filtrate flux was performed based on the tests conducted with the same test matrix on the bench- and lab-scale units. A linear scale-up pattern was observed with respect to the number of discharge ports for flux values less than 30 L/hr/m2, with the flux of the lab-scale unit being twice the flux of the bench-scale unit. For higher values of flux, the scale-up relationship transformed to a logarithmic function. An overall logarithmic model, with an R2 value of 0.85, was developed to fit the fluxes from the bench-scale unit with the lab-scale unit. The variation in cake moisture, defined as the residual moisture content in the solid stream, as a function of time was shown to follow the compressional rheology model of compaction and consolidation. Results from a fractional factorial design on the lab-scale unit showed that the cake moisture at pH ~3 was about half of what was observed at pH ~12, indicating that pH had a significant effect on the final cake moisture. The zeta potential at acidic pH had values greater than -15 mV, which indicated the presence of agglomeration resulting in improved filtration performance at pH ~3. An inverse relationship between feed solids concentration and final cake moisture was established as test results showed that a concentrated feed provided a drier cake. Testing with a Box-Behnken design showed a drop in cake moisture by 13% with an increase in filtration time from 40 minutes to 50 minutes. The response surface design also showed a 30% reduction in cake moisture for a test with 10 minutes of air-blow after 30 minutes of filtration time compared to a test with 40 minutes of filtration time. A regression model for the response surface with respect to filtration time, air-blow time, and pH had an R2 value of about 0.99 with almost all the main effects, their square terms, and the interaction effects statistically significant. The ANOVA analysis based on a split-plot design to study the effects of cake thickness and pressure showed that both factors significantly affected the cake moisture. A regression model was developed with an R2 value of about 0.99 to relate cake thickness, the single largest statistically significant factor, with the weight of solids in the cake. For a product specification of 25% cake moisture, it was observed that the thinnest cake had the highest unit capacity in terms of the solids deposited per unit time and the amount of filtrate recovered per unit filter volume. Filtrate flux values increased with increase in cake thickness, and the flux values corrected for cake thickness were similar for cakes with thickness between 30 and 40 mm. A full-factorial design to study the effect of temperature showed that at about 50 minutes, the cake moisture for a test at 43 oC was about half as much as the cake moisture obtained for the test conducted at 10 oC, thereby establishing temperature as a variable of significance. To analyze the effects of pressure, pH, temperature, solids concentration, filtration time, air-blow time, and cake thickness on cake moisture, a 9-factor regression model based on an exhaustive search algorithm and a 7-6-1 ANN based on a resilient backpropagation algorithm were developed with R2 values of 0.84 and 0.94, respectively. Relative importance of each input variable was analyzed using novel methods such as added-variable plots based on the regression model and Olden’s method based on the ANN model. Results from both methods established a negative correlation for pressure, solids concentration, filtration time, temperature, and air-blow time and a positive correlation for cake thickness and pH. Even though both models served as good interpretable models, it was shown that the ANN model outperformed the regression model in terms of predictive capability, with an R2 value of 0.965 compared with the regression model’s 0.750 for a test dataset. An attempt was made to use zeolite-based additives to capture and retain metals leaching from refuse stream at acidic pH. The additives were shown to have minimal effect on the filtration performance, while adsorbing as high as 80% of iron. The adsorption was not significantly affected by slight variations in initial metal concentration. The treatment was shown to retain most of the metals when the treated cakes were exposed to a water with a pH ~5, comparable to that of rain. In summary, this study evaluated the effects of fines fraction, pressure, slurry pH, solids concentration, temperature, cake thickness, filtration time, and air-blow time on pressure filtration of coal refuse slurry using statistically designed experiments. Bench-scale testing indicated that filtration performance, measured in terms of cumulative filtrate flux, increased with pressure and temperature and decreased with fines fraction, pH, solids concentration, and time. Experimental results and statistical analyses indicated improved filtration performance, measured in terms of cake moisture, in the lab-scale unit with increase in pressure, solids concentration, temperature, filtration time, and air-blow time, while the performance decreased with increase in pH and cake thickness. Modeling was performed based on the experimental data using linear regression and ANN. The ANN model was shown to have a better predictive performance compared to the regression model. Based on the experimental data and the sensitivity analyses performed using the models, pH was identified as a major factor that influenced the filtration performance. Also, zeolite-based additives were shown to reduce iron leaching out of refuse at lower pH values when mixed with the slurry prior to pressure filtration.