Carbothermal reduction of phosphogypsum waste using coal

Abstract

Phosphogypsum waste represents a potential source for the recovery of elemental sulphur, rare earth elements, and calcium carbonate. With primary interest on the reducing step of the process, converting the sulphates to sulphides, there is good precedence shown that the reduction of phosphogypsum can be realised carbothermally. Coal should serve as a suitable source of carbon for the process step, with the ready availability of coal and relatively low cost being key factors in its choice. Calcium sulphate is reduced within a temperature range of 850–1100 °C, with an unfavourable side reaction, that produces calcium oxide, that can initiate at 900 °C depending on process conditions. The side reaction is an oxidative reaction, and its negative impact can thus be mitigated by charging sufficient excess carbon which serves to maintain the reducing conditions required for the primary reaction. The reduction is generally believed to be facilitated by a gaseous intermediate acting as the reducing agent. The presence of oxygen and carbon dioxide are undesirable as they favour the production of calcium oxide and calcium sulphate rather than the sulphide. It is recommended that, for the use of coal as carbonaceous material, an inert environment (such as a nitrogen atmosphere) should be created to carry out the reduction. The addition of catalysts such as ferric oxide or potassium dichromate can enhance the reaction satisfactorily and reduce the initiation temperature to approximately 750 °C and increase the yield of calcium sulphide. Heat transfer within the reacting bed of phosphogypsum and coal is modelled by means of a two-dimensional transient finite difference method. The model’s scope has been limited from describing the entire furnace and all the involved conductive, convective, and radiant heat transfer to rather focus on the heat transfer within the reacting material contained in the furnace. The kinetics of the reaction is used to develop a relationship between the temperature of a node in the two-dimensional mesh and the correlating composition of said node. This change in composition is incorporated into the model by adjusting the heat transfer properties (heat capacity, alpha, tau) used in the heat transfer calculation for each node respective to its temperature. The model is validated against experimental data, captured using an array of thermocouples placed throughout the reacting mixture in a three-inch and six-inch crucible. The vessels were heated to 1000 °C with a heating rate of 3 °C per min while purging with nitrogen gas. The model achieves conservative results, over-predicting the time required to reach 1000 °C. It can be noted that improvement in model accuracy can be achieved by varying the pre-exponential factor and the thermal conductivity of the system. However, the effect the thermal conductivity has is far more substantial than the rate of reaction for the system. This indicates that the thermal conductivity values available from literature (which are limited) do not describe the system with sufficient accuracy and as such, it would be beneficial to study the thermal conductivity. Application of a model that varies thermal conductivity linearly with temperature achieves reasonable results and matches the trends observed in experimental results well. However, it is still recommended that larger scale experiments be conducted as the model shows increasing inaccuracy as the mass and vessel diameter increase. A sensitivity analysis for a hypothetical tunnel kiln is used to determine the optimum cylindrical vessel configuration to maximise the possible material processing rate. The analysis found that a shallower material bed results in a more optimal setup, with more efficient heat transfer. This results in an estimated processing rate 771 Mt per year, which is not unrealistic based on similar tunnel kilns used in the direct reduction of iron. However, the design of a furnace is a complicated process with many variables, so these results are far from conclusive and it is recommended to determine whether a shallow square/cubic configuration is better suited. The model has good flexibility and can be updated and improved readily. Conservative results indicate a good applicability for the model in industrial scenarios where factors of safety are necessary. The model shows good suitability for the prediction of the heat transfer occurring within a reacting bed of solid phosphogypsum and coal.