Publication

Abstract : The conventional laboratory testing of concrete for its fresh stage property is time-consuming, expensive-causing a loss of resources, materials. However, nowadays artificial intelligence-based robust systems are very well used in many engineering disciplines. As cement paste workability has a very good influence on concrete workability, a Machine-Learning model like SVM (Support Vector Machine) is calibrated, validated, and finally established to predict the Marsh Cone flow time of cement paste by giving ingredient property as the inputs. This research proposes an intelligent method for predicting the Marsh Cone flow time of cement paste based on machine learning methods. The experiment was carried out on a superplasticized cement paste made with locally available brands of Portland pozzolana cement (PPC) and superplasticizers from four different families at a water-cement ratio of 0.37. The four different families namely PCE (Polycarboxylate Ether), SNF (Sulfonated naphthalene-formaldehyde condensates), SMF (Sulfonated melamine–formaldehyde condensates), and (Lignosulphonates) were used. The model used water, cement, and the amount of superplasticizer (which was divided into seven different inputs based on their family and brand) as input parameters, with marsh cone flow time as the output parameter. A marsh cone flow test on more than 200 superplasticized cement paste mixes was used to obtain the model’s testing and training results. The model was found to reliably predict the flow time of the Portland Pozzolana Cement paste within the experimental domain when the expected and measured values of marsh cone flow time were compared. The prediction capacity of SVM is checked with that of KNN (K-nearest neighbors algorithm) and Random forest and it is found that prediction accuracy is more for SVM compared to other models.