Manufacture of artificial lightweight aggregates recycled from anaerobic digested sewage sludge and process optimization by machine learning modeling techniques

In this study, in an effort to manufacture lightweight aggregates recycled from anaerobic digested sewage sludge (ADSS), the firing process characteristics of artificial lightweight aggregates containing a large amount of ADSS were identified and the manufacturing process was optimized and modeled using machine learning techniques. Samples were prepared by rapid firing and according to the orthogonal arrangement experimental design, the particle density and water absorption rate were measured, and a pore analysis was conducted by 3D-CT and the mercury intrusion method. Random forest, SVR, and XGBoost techniques were used as machine learning techniques to analyze the data. As a result of the rapid firing test, samples with an ADSS content of 40 wt% or more were found to have a low density and were found not to show bloating. For aggregates containing more than 40% ADSS, a loss of organic matter played a greater role in the decreased density than bloating due to viscous behavior. Data expansion using an orthogonal arrangement table and modeling by the SVR method for the ADSS 50 mixture found an RMSE value of 0.013 with significant predicted values of the physical properties also obtained. Aggregates with a content of 50 wt% of anaerobically digested sludge were successfully prepared on a pilot scale, and this aggregate mix met EN-13055 standards.