Model Prediction and Multi-Objective Optimization of Unfired Bricks Incorporated with Drinking Water Treatment Sludge Using Machine Learning

Incorporating drinking water treatment sludge (DWTS) into unfired bricks provides a promising approach for large-scale utilization with low carbon emission. However, the complex effects of material composition and curing strategy on unfired brick performance are difficult to optimize through conventional trial-and-error methods. Therefore, in this study machine learning (ML) combined with Pareto front analysis was introduced to develop a multi-objective optimization of both compressive strength and cost. Among the tested models, the random forest (RF) and extreme gradient boosting (XGBoost) exhibited the best generalization performance for predicting compressive strength and cost based on the 5-fold cross validation, respectively. SHapley Additive exPlanation (SHAP) analysis revealed that early water immersion followed by standard curing strongly enhanced compressive strength and DWTS proportion had the greatest negative influence on cost. Pareto optimization identified a trade-off scheme with the predicted compressive strength of 15.5 MPa and a negative cost of −2.4 RMB. The measured compressive strength of this optimal sample was 15.08 MPa, close to the predicted value and much higher than that of the reference sample. Scanning electron microscopy (SEM) and thermogravimetry analysis (TGA) results further confirmed abundant hydration products in the optimal sample. This study highlights the potential of ML to guide DWTS utilization in unfired bricks while balancing compressive strength and cost.