Machine learning prediction models for compressive strength of calcined sludge-cement composites

Replacing part of ordinary Portland cement with sludge can reduce the use of cement while recycling sludge and achieve low CO2 emissions, which is an environment-friendly method for sludge treatment. However, compre-hensive research on compressive strength of calcined sludge-cement composites has not formed due to numerous influencing factors. In this paper, experiments are designed under six factors to study the effects of various factors on the compressive strength of calcined sludge-cement composites, involving ball milling time, calcination temperature, sludge replacement rate, curing age, admixtures amount of calcium chloride (CaCl2) and calcium sulfate (CaSO4). At the same time, this paper uses machine learning to establish six different regression pre-diction models to predict the compressive strength, including Linear Regression (LR), Support Vector Machine Regression (SVR), Random Forest Regression (RF), Multi-layer Perceptron-Artificial Neural Network Regression (MLP-ANN), Ensemble Regression and Convolutional Neural Network Regression (CNN). According to the re-sults, CNN and Ensemble Regression models provide the excellent prediction accuracy. By comparing the robustness, curing age has the greatest impact on compressive strength, while the influence of ball milling time and CaSO4 is small, which is consistent with the experimental results.

Automated summary

This study investigates the effects of various factors on the compressive strength of calcined sludge-cement composites through experiments, and uses machine learning to establish six different regression prediction models to predict compressive strength. The results show that CNN and Ensemble Regression models provide excellent prediction accuracy, with curing age having the greatest impact on compressive strength.