Modelling a linear flow tubular reactor for high sulphate water treatment

Abstract

The increased abandonment of mines in South Africa, along with current acid mine drainage (AMD) production from operating mines, continues to pollute clean water resources. Various methods exist to treat AMD, including passive and active options. Biological sulphate reduction (BSR) is a process that employs sulphate-reducing bacteria to treat wastewater by removing sulphate and heavy metals and producing sulphides. This process requires a carbon source, which acts as an electron donor for the bacteria. A 12 L linear flow channel reactor (LFCR) was designed and used for this study. Ethanol was selected as the carbon source. AMD and the inoculum (containing SRB) were mixed with ethanol and fed into the reactor at a flow rate of 84 mg/L. The reactor operated for 100 days, maintaining constant parameters with respect to temperature, feed flow rate, and ethanol volume. Variables such as the AMD/inoculum ratio, hydraulic residence time (HRT), and sulphate/COD ratio were adjusted. The operation was divided into three periods: Period I (40 days), Period II (30 days), and Period III (40 days). Period I showed better performance compared to periods II and III. This performance was due to SRB adapting to the reactor conditions and the inoculum being highly active. During this period, the pH remained neutral between 7 and 7.5. Heavy metal removal exceeded 95%, sulphate removal reached 99%, COD reduction was 95%, and sulphide production was 195 mg/L. Period II exhibited the poorest reactor performance due to the higher AMD/inoculum ratio, resulting in improved acidic conditions in the reactor. Although Period III demonstrated good performance, the biofilm's degradation prevented it from reaching the same level as that of Period I. Variations in HRT and the sulphate/COD ratio also influenced the reactor's performance. The combined effect of lowering the HRT and increasing the sulphate/COD ratio resulted in reduced sulphate removal efficiency. Overall, the reactor performed well in treating AMD. The reactor's performance in sulphate removal was modelled using linear regression and an artificial neural network (ANN). The linear regression model's performance was as follows: test R2 was 0.42, root mean square error (RMSE) was 735 and mean absolute error (MAE) was 541. The ANN's performance was as follows: test R2 was 0.75, RMSE was 483, and MAE was 305. The ANN significantly outperformed the linear regression. Adding more data points could enhance the accuracy of both models.